{"schema":"5lc-index-ledger/1","key":"models-data-evaluation","layer":"models","sublayer":"data-evaluation","asOf":"2026-10-22","about":"5LC Importance Index ledger for one sublayer: the measure card, the current ten with any editorial overrides and their reasons, the computed result for every scored company (Importance = share of AI output × years to replace, Reach, replacement years, Rising, the Monte Carlo rank simulation and its stability class, validation, the revenue baseline and rater agreement), each scored company's sourced inputs and irreplaceability ratings, and the sublayer's validation cases with their sources. Monte Carlo frequencies describe the stated input ranges, not true probabilities. Omitted: the free-text notes of the internal check passes and of rating reconciliations, and the location of the internal card review; references to local evidence files and to internal method decisions are replaced by plain descriptions.","roster":{"members":[{"entityId":"mercor","slot":"importance","since":"2026-10-22"},{"entityId":"surge-ai","slot":"importance","since":"2026-10-22"},{"entityId":"handshake-ai","slot":"importance","since":"2026-10-22"},{"entityId":"micro1","slot":"importance","since":"2026-10-22"},{"entityId":"scale-ai","slot":"importance","since":"2026-10-22"},{"entityId":"arena-ai","slot":"importance","since":"2026-10-22"},{"entityId":"snorkel-ai","slot":"importance","since":"2026-10-22"},{"entityId":"turing","slot":"importance","since":"2026-10-22"},{"entityId":"innodata","slot":"importance","since":"2026-10-22"},{"entityId":"proximal","slot":"rising","since":"2026-10-22"}],"pending":{},"lastRun":"2026-10-22","resultsDate":"2026-10-22"},"card":{"schema":"5lc-index-card/1","key":"models-data-evaluation","layer":"models","sublayer":"data-evaluation","version":1,"scope":"Human data, training tasks and tests sold to model developers: expert data for post-training, large-scale labelling and multimodal data, reinforcement-learning environments, independent model evaluation, and labelling, data-curation and synthetic-data software; content licensing by publishers and platforms is outside this sublayer, evaluation and monitoring tools for companies' own AI applications count in Applications · Data platforms, AI applications that data vendors sell to enterprises and governments count in Applications, and labs' in-house data teams are not counted.","outputMeasure":"Revenue from the data, training tasks and tests that model developers buy.","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","hardToReplace":"Labs use several data vendors at once and can move work within months. Harder to replace are large pools of vetted experts, environment builders with particular skills, and community evaluations with years of comparison data.","functions":[{"id":"expert-data","name":"Expert data for post-training","why":"Specialists write answers, grade outputs and build reasoning examples; most of the spending.","weight":{"low":0.44,"central":0.629,"high":0.805},"weightSource":{"url":"https://techcrunch.com/2026/07/08/these-ai-startups-are-growing-revenue-at-faster-and-faster-rates/","title":"These AI startups are growing revenue at faster and faster rates","publisher":"TechCrunch","publishedDate":"2026-07-08","accessedDate":"2026-10-22","tier":"T2","locator":"The company says (CEO Brendan Foody): Mercor crossed $2 billion in gross annualized revenue as of June 2026, four months after reaching $1 billion"},"shareData":"Company run-rate statements and press reports, split by each vendor's services"},{"id":"labelling-multimodal","name":"Large-scale labelling and multimodal data","why":"High-volume labelling of images, video, speech and sensor data.","weight":{"low":0.068,"central":0.155,"high":0.281},"weightSource":{"url":"https://csv-storage.forexpros.com/slides/c7ad004a2be7f7acb476cfc22640135f657a661c0d33df530e0bb4da3a481d97.pdf","title":"TaskUs Investor Presentation - Q2 2026","publisher":"TaskUs (slides hosted by Investing.com)","publishedDate":"2026-08-05","accessedDate":"2026-10-22","tier":"T2","locator":"Service-line slide: AI Services TTM 6/26 revenue $244mm (20% of revenue), covering autonomous vehicles, generative AI, object identification and classification, advanced robotics and OCR; AI Services grew 25.8% year over year in Q2 2026"},"shareData":"Filings of listed vendors (TaskUs, Appen, Innodata, DataOcean AI) and company statements"},{"id":"rl-environments","name":"RL environments and agent tasks","why":"Simulated software and graded tasks for training agents; a newer part of lab spending.","weight":{"low":0.059,"central":0.147,"high":0.272},"weightSource":{"url":"https://epoch.ai/gradient-updates/state-of-rl-envs","title":"An FAQ on reinforcement learning environments","publisher":"Epoch AI","publishedDate":"2026-01-12","accessedDate":"2026-10-22","tier":"T4","locator":"Cites The Information (September 2025): Anthropic had discussed spending over $1 billion on RL environments over the following year; contracts often six to seven figures per quarter; Mercor, Surge, Handshake and Turing sell environments"},"shareData":"Environment contracts in company statements and press reports (labelled estimates)"},{"id":"evaluation","name":"Independent model evaluation","why":"Leaderboards, expert test sets and red-teaming that labs pay for.","weight":{"low":0.022,"central":0.045,"high":0.079},"weightSource":{"url":"https://arena.ai/blog/arena-100m-revenue","title":"Arena Reaches $100M in 8 Months","publisher":"Arena","publishedDate":"2026-06-29","accessedDate":"2026-10-22","tier":"T2","locator":"Arena crossed $100M annualized revenue run rate within eight months of launching its enterprise offering; 10M+ monthly visitors, 82M+ votes"},"shareData":"Company run-rate statements and press reports"},{"id":"data-tooling","name":"Labelling and data software","why":"Software for labelling, curating and generating training data; small next to the services.","weight":{"low":0.008,"central":0.024,"high":0.056},"weightSource":{"url":"https://snorkel.ai/blog/data-2-0-and-the-research-era-of-ai-data/","title":"Data 2.0 and the research era of AI data","publisher":"Snorkel AI","publishedDate":"2026-09-22","accessedDate":"2026-10-22","tier":"T2","locator":"Snorkel crossed a $375M annualized revenue run rate; it has grown over 18x since launching its data-as-a-service offering nearly a year earlier, so its software platform is a small part"},"shareData":"Software revenue statements of labelling-platform companies"}],"risingFloorPct":1,"validationCaseIds":[],"knownGaps":["No vendor discloses its revenue split across these functions; the weights split each vendor's run-rate by the services it describes (estimates), and the ranges allow each vendor's smaller parts to be half or one and a half times the stated split.","Most vendors are private: run-rates are company statements or press reports, and some publish gross figures before contributor pay while others publish net (micro1 gave both in August 2026).","Surge AI and Scale AI have not published current revenue; their sizes are estimates, and Scale's enterprise and government applications are left out.","Units of larger groups (TELUS Digital, EXL's iMerit, Uber AI Solutions) are measured from their own AI-data revenue, never the parent's; TELUS Digital does not report it separately.","Labs' in-house data teams and direct contractor hiring are not measured, so the totals understate lab spending on data."],"approval":{"approver":"Claude (delegated by Dachi)","date":"2026-10-22","sha256":"b68aea93ce4e49c91765aa53b83cb90cc3ed62da4106fe38fe35eb4c01f94de5","delegatedOn":"2026-10-22"}},"result":{"schema":"5lc-index-result/1","key":"models-data-evaluation","asOf":"2026-10-22","methodVersion":"1.1","methodSha256":"3cabed6b38514c263d8b86a494e8cc19cdbd0b2a1013543dc024192c49f7a6a6","cardSha256":"b68aea93ce4e49c91765aa53b83cb90cc3ed62da4106fe38fe35eb4c01f94de5","validationSha256":"409931bb458f38b02342c9bf43e61eab299e47d8679912af2c96074ec556f17e","inputsSha256":"bb8dee13d34b79a83fc271157de4600d950ad389fa9b5cfa68d70a8010570b4d","unscreened":[],"companies":[{"entityId":"mercor","name":"Mercor","reachPct":19.026151,"importance":9.17527761927,"importanceRank":1,"functions":[{"id":"expert-data","weight":0.629,"sharePct":23.291,"shareLowPct":18.7,"shareHighPct":31.074,"stale":false,"offBasis":false,"level":2,"years":0.5,"disagreement":false},{"id":"labelling-multimodal","weight":0.155,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"rl-environments","weight":0.147,"sharePct":26.321,"shareLowPct":13.97,"shareHighPct":40.547,"stale":false,"offBasis":false,"level":1.83333333333,"years":0.4454493590691404,"disagreement":false},{"id":"evaluation","weight":0.045,"sharePct":11.265,"shareLowPct":5.456,"shareHighPct":19.508,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false},{"id":"data-tooling","weight":0.024,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false}],"rising":88.6363636363,"risingEligible":true,"risingComponents":{"growth":77.2727272727,"shareChange":100},"mc":{"medianRank":1,"rankP10":1,"rankP90":2,"pTop8":1,"pSelected":1,"stability":"firm"},"baselineRevenueRank":1},{"entityId":"surge-ai","name":"Surge AI","reachPct":15.507774,"importance":6.61197427425,"importanceRank":2,"functions":[{"id":"expert-data","weight":0.629,"sharePct":18.294,"shareLowPct":5.4492,"shareHighPct":34.4888,"stale":true,"offBasis":false,"level":1.83333333333,"years":0.4454493590691404,"disagreement":false},{"id":"labelling-multimodal","weight":0.155,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"rl-environments","weight":0.147,"sharePct":22.514,"shareLowPct":2.9652,"shareHighPct":46.5788,"stale":true,"offBasis":false,"level":1.66666666667,"years":0.3968502629929668,"disagreement":false},{"id":"evaluation","weight":0.045,"sharePct":15.362,"shareLowPct":1.7276,"shareHighPct":34.2844,"stale":true,"offBasis":false,"level":1,"years":0.25,"disagreement":false},{"id":"data-tooling","weight":0.024,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false}],"rising":null,"risingEligible":true,"risingComponents":{},"mc":{"medianRank":2,"rankP10":1,"rankP90":2,"pTop8":1,"pSelected":1,"stability":"firm"},"baselineRevenueRank":4},{"entityId":"handshake-ai","name":"Handshake AI","reachPct":9.164196,"importance":3.13538784449,"importanceRank":3,"functions":[{"id":"expert-data","weight":0.629,"sharePct":12.45,"shareLowPct":10.916,"shareHighPct":16.166,"stale":false,"offBasis":false,"level":1.5,"years":0.3535533905932738,"disagreement":false},{"id":"labelling-multimodal","weight":0.155,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"rl-environments","weight":0.147,"sharePct":7.423,"shareLowPct":3.678,"shareHighPct":12.961,"stale":false,"offBasis":false,"level":1.16666666667,"years":0.2806155120779916,"disagreement":false},{"id":"evaluation","weight":0.045,"sharePct":5.377,"shareLowPct":2.634,"shareHighPct":9.545,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false},{"id":"data-tooling","weight":0.024,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false}],"rising":74.3506493507,"risingEligible":true,"risingComponents":{"growth":77.2727272727,"shareChange":71.4285714286},"mc":{"medianRank":3,"rankP10":3,"rankP90":4,"pTop8":1,"pSelected":1,"stability":"firm"},"baselineRevenueRank":2},{"entityId":"micro1","name":"micro1","reachPct":8.739662,"importance":2.99822461758,"importanceRank":4,"functions":[{"id":"expert-data","weight":0.629,"sharePct":11.998,"shareLowPct":5.852,"shareHighPct":15.114,"stale":false,"offBasis":false,"level":1.5,"years":0.3535533905932738,"disagreement":false},{"id":"labelling-multimodal","weight":0.155,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"rl-environments","weight":0.147,"sharePct":7.07,"shareLowPct":1.866,"shareHighPct":12.044,"stale":false,"offBasis":false,"level":1.16666666667,"years":0.2806155120779916,"disagreement":false},{"id":"evaluation","weight":0.045,"sharePct":3.414,"shareLowPct":0.876,"shareHighPct":5.981,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false},{"id":"data-tooling","weight":0.024,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false}],"rising":81.4935064935,"risingEligible":true,"risingComponents":{"growth":77.2727272727,"shareChange":85.7142857143},"mc":{"medianRank":4,"rankP10":3,"rankP90":5,"pTop8":1,"pSelected":1,"stability":"firm"},"baselineRevenueRank":3},{"entityId":"scale-ai","name":"Scale AI","reachPct":6.560824,"importance":1.79048942694,"importanceRank":5,"functions":[{"id":"expert-data","weight":0.629,"sharePct":5.823,"shareLowPct":2.84,"shareHighPct":10.407,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false},{"id":"labelling-multimodal","weight":0.155,"sharePct":14.921,"shareLowPct":6.555,"shareHighPct":25.967,"stale":false,"offBasis":false,"level":1.33333333333,"years":0.3149802624729905,"disagreement":false},{"id":"rl-environments","weight":0.147,"sharePct":1.631,"shareLowPct":0.659,"shareHighPct":3.211,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false},{"id":"evaluation","weight":0.045,"sharePct":7.681,"shareLowPct":3.221,"shareHighPct":14.266,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false},{"id":"data-tooling","weight":0.024,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false}],"rising":null,"risingEligible":true,"risingComponents":{"growth":0},"mc":{"medianRank":5,"rankP10":4,"rankP90":5,"pTop8":1,"pSelected":1,"stability":"firm"},"baselineRevenueRank":5},{"entityId":"arena-ai","name":"Arena","reachPct":0.844875,"importance":0.844875,"importanceRank":6,"functions":[{"id":"expert-data","weight":0.629,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"labelling-multimodal","weight":0.155,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"rl-environments","weight":0.147,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"evaluation","weight":0.045,"sharePct":18.775,"shareLowPct":17.365,"shareHighPct":23.966,"stale":false,"offBasis":false,"level":3,"years":1,"disagreement":false},{"id":"data-tooling","weight":0.024,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false}],"rising":null,"risingEligible":false,"risingComponents":{},"mc":{"medianRank":9,"rankP10":6,"rankP90":15,"pTop8":0.4992,"pSelected":0.5714,"stability":"borderline"},"baselineRevenueRank":14},{"entityId":"snorkel-ai","name":"Snorkel AI","reachPct":3.282215,"importance":0.82055375,"importanceRank":7,"functions":[{"id":"expert-data","weight":0.629,"sharePct":3.97,"shareLowPct":3.33,"shareHighPct":4.893,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false},{"id":"labelling-multimodal","weight":0.155,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"rl-environments","weight":0.147,"sharePct":2.039,"shareLowPct":1.03,"shareHighPct":3.223,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false},{"id":"evaluation","weight":0.045,"sharePct":3.2,"shareLowPct":1.626,"shareHighPct":5.024,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false},{"id":"data-tooling","weight":0.024,"sharePct":14.223,"shareLowPct":7.656,"shareHighPct":20.968,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false}],"rising":67.2077922078,"risingEligible":true,"risingComponents":{"growth":77.2727272727,"shareChange":57.1428571429},"mc":{"medianRank":8,"rankP10":7,"rankP90":11,"pTop8":0.5416,"pSelected":1,"stability":"firm"},"baselineRevenueRank":6},{"entityId":"turing","name":"Turing","reachPct":3.01222,"importance":0.777038059388,"importanceRank":8,"functions":[{"id":"expert-data","weight":0.629,"sharePct":3.458,"shareLowPct":1.583,"shareHighPct":6.939,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false},{"id":"labelling-multimodal","weight":0.155,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"rl-environments","weight":0.147,"sharePct":5.329,"shareLowPct":1.583,"shareHighPct":12.645,"stale":false,"offBasis":false,"level":1.16666666667,"years":0.2806155120779916,"disagreement":false},{"id":"evaluation","weight":0.045,"sharePct":1.195,"shareLowPct":0.344,"shareHighPct":3.016,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false},{"id":"data-tooling","weight":0.024,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false}],"rising":null,"risingEligible":true,"risingComponents":{"growth":9.09090909091},"mc":{"medianRank":7,"rankP10":6,"rankP90":13,"pTop8":0.6331,"pSelected":0.6975,"stability":"likely"},"baselineRevenueRank":9},{"entityId":"innodata","name":"Innodata","reachPct":2.883725,"importance":0.747633640402,"importanceRank":9,"functions":[{"id":"expert-data","weight":0.629,"sharePct":2.795,"shareLowPct":1.488,"shareHighPct":4.121,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false},{"id":"labelling-multimodal","weight":0.155,"sharePct":5.627,"shareLowPct":2.295,"shareHighPct":9.114,"stale":false,"offBasis":false,"level":1.16666666667,"years":0.2806155120779916,"disagreement":false},{"id":"rl-environments","weight":0.147,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"evaluation","weight":0.045,"sharePct":5.633,"shareLowPct":2.297,"shareHighPct":9.123,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false},{"id":"data-tooling","weight":0.024,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false}],"rising":29.8701298701,"risingEligible":true,"risingComponents":{"growth":45.4545454545,"shareChange":14.2857142857},"mc":{"medianRank":10,"rankP10":7,"rankP90":13,"pTop8":0.2469,"pSelected":0.3593,"stability":"borderline"},"baselineRevenueRank":7},{"entityId":"prolific","name":"Prolific","reachPct":2.623506,"importance":0.687037450899,"importanceRank":10,"functions":[{"id":"expert-data","weight":0.629,"sharePct":2.964,"shareLowPct":1.342,"shareHighPct":5.822,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false},{"id":"labelling-multimodal","weight":0.155,"sharePct":3.411,"shareLowPct":0.991,"shareHighPct":8.111,"stale":false,"offBasis":false,"level":1.16666666667,"years":0.2806155120779916,"disagreement":false},{"id":"rl-environments","weight":0.147,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"evaluation","weight":0.045,"sharePct":5.121,"shareLowPct":1.506,"shareHighPct":11.888,"stale":false,"offBasis":false,"level":1.33333333333,"years":0.3149802624729905,"disagreement":false},{"id":"data-tooling","weight":0.024,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false}],"rising":null,"risingEligible":true,"risingComponents":{},"mc":{"medianRank":9,"rankP10":6,"rankP90":14,"pTop8":0.3967,"pSelected":0.4949,"stability":"borderline"},"baselineRevenueRank":8},{"entityId":"appen","name":"Appen","reachPct":2.385366,"importance":0.643681654016,"importanceRank":11,"functions":[{"id":"expert-data","weight":0.629,"sharePct":1.334,"shareLowPct":0.597,"shareHighPct":2.204,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false},{"id":"labelling-multimodal","weight":0.155,"sharePct":9.976,"shareLowPct":6.715,"shareHighPct":13.515,"stale":false,"offBasis":false,"level":1.16666666667,"years":0.2806155120779916,"disagreement":false},{"id":"rl-environments","weight":0.147,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"evaluation","weight":0.045,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"data-tooling","weight":0.024,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false}],"rising":9.0909090909,"risingEligible":true,"risingComponents":{"growth":18.1818181818,"shareChange":0},"mc":{"medianRank":12,"rankP10":8,"rankP90":15,"pTop8":0.1301,"pSelected":0.1745,"stability":"borderline"},"baselineRevenueRank":11},{"entityId":"taskus","name":"TaskUs","reachPct":2.25347,"importance":0.602725883862,"importanceRank":12,"functions":[{"id":"expert-data","weight":0.629,"sharePct":1.445,"shareLowPct":0.694,"shareHighPct":2.391,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false},{"id":"labelling-multimodal","weight":0.155,"sharePct":8.294,"shareLowPct":5.095,"shareHighPct":12.178,"stale":false,"offBasis":false,"level":1.16666666667,"years":0.2806155120779916,"disagreement":false},{"id":"rl-environments","weight":0.147,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"evaluation","weight":0.045,"sharePct":1.311,"shareLowPct":0.629,"shareHighPct":2.17,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false},{"id":"data-tooling","weight":0.024,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false}],"rising":27.9220779221,"risingEligible":true,"risingComponents":{"growth":27.2727272727,"shareChange":28.5714285714},"mc":{"medianRank":13,"rankP10":9,"rankP90":16,"pTop8":0.0752,"pSelected":0.1068,"stability":"borderline"},"baselineRevenueRank":10},{"entityId":"invisible-technologies","name":"Invisible Technologies","reachPct":2.192278,"importance":0.5480695,"importanceRank":13,"functions":[{"id":"expert-data","weight":0.629,"sharePct":2.117,"shareLowPct":0.3206,"shareHighPct":5.8774,"stale":true,"offBasis":false,"level":1,"years":0.25,"disagreement":false},{"id":"labelling-multimodal","weight":0.155,"sharePct":4.263,"shareLowPct":0.2924,"shareHighPct":12.6376,"stale":true,"offBasis":false,"level":1,"years":0.25,"disagreement":false},{"id":"rl-environments","weight":0.147,"sharePct":1.36,"shareLowPct":0.085,"shareHighPct":4.243,"stale":true,"offBasis":false,"level":1,"years":0.25,"disagreement":false},{"id":"evaluation","weight":0.045,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"data-tooling","weight":0.024,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false}],"rising":null,"risingEligible":true,"risingComponents":{},"mc":{"medianRank":11,"rankP10":6,"rankP90":16,"pTop8":0.3061,"pSelected":0.3687,"stability":"borderline"},"baselineRevenueRank":13},{"entityId":"telus-digital","name":"TELUS Digital","reachPct":1.820557,"importance":0.488480461118,"importanceRank":14,"functions":[{"id":"expert-data","weight":0.629,"sharePct":1.163,"shareLowPct":0.305,"shareHighPct":2.855,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false},{"id":"labelling-multimodal","weight":0.155,"sharePct":7.026,"shareLowPct":2.55,"shareHighPct":14.366,"stale":false,"offBasis":false,"level":1.16666666667,"years":0.2806155120779916,"disagreement":false},{"id":"rl-environments","weight":0.147,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"evaluation","weight":0.045,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"data-tooling","weight":0.024,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false}],"rising":null,"risingEligible":true,"risingComponents":{},"mc":{"medianRank":14,"rankP10":9,"rankP90":18,"pTop8":0.0757,"pSelected":0.1001,"stability":"borderline"},"baselineRevenueRank":null},{"entityId":"rws-trainai","name":"RWS TrainAI","reachPct":1.766253,"importance":0.479327177993,"importanceRank":15,"functions":[{"id":"expert-data","weight":0.629,"sharePct":0.847,"shareLowPct":0.256,"shareHighPct":1.762,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"labelling-multimodal","weight":0.155,"sharePct":7.958,"shareLowPct":3.916,"shareHighPct":12.815,"stale":false,"offBasis":false,"level":1.16666666667,"years":0.2806155120779916,"disagreement":false},{"id":"rl-environments","weight":0.147,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"evaluation","weight":0.045,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"data-tooling","weight":0.024,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false}],"rising":null,"risingEligible":true,"risingComponents":{"growth":36.3636363636},"mc":{"medianRank":15,"rankP10":11,"rankP90":18,"pTop8":0.0311,"pSelected":0.0413,"stability":"borderline"},"baselineRevenueRank":null},{"entityId":"proximal","name":"Proximal","reachPct":1.651874,"importance":0.447230656337,"importanceRank":16,"functions":[{"id":"expert-data","weight":0.629,"sharePct":0.847,"shareLowPct":0.256,"shareHighPct":1.576,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"labelling-multimodal","weight":0.155,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"rl-environments","weight":0.147,"sharePct":7.613,"shareLowPct":3.74,"shareHighPct":11.118,"stale":false,"offBasis":false,"level":1.16666666667,"years":0.2806155120779916,"disagreement":false},{"id":"evaluation","weight":0.045,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"data-tooling","weight":0.024,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false}],"rising":60.0649350649,"risingEligible":true,"risingComponents":{"growth":77.2727272727,"shareChange":42.8571428571},"mc":{"medianRank":16,"rankP10":12,"rankP90":20,"pTop8":0.0122,"pSelected":1,"stability":"firm"},"baselineRevenueRank":12},{"entityId":"labelbox","name":"Labelbox","reachPct":1.317734,"importance":0.3294335,"importanceRank":17,"functions":[{"id":"expert-data","weight":0.629,"sharePct":1.27,"shareLowPct":0.342,"shareHighPct":4.375,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false},{"id":"labelling-multimodal","weight":0.155,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"rl-environments","weight":0.147,"sharePct":0.816,"shareLowPct":0.164,"shareHighPct":3.185,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"evaluation","weight":0.045,"sharePct":1.28,"shareLowPct":0.259,"shareHighPct":4.931,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false},{"id":"data-tooling","weight":0.024,"sharePct":14.223,"shareLowPct":3.21,"shareHighPct":39.878,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false}],"rising":null,"risingEligible":true,"risingComponents":{"growth":77.2727272727},"mc":{"medianRank":15,"rankP10":10,"rankP90":20,"pTop8":0.0397,"pSelected":0.0686,"stability":"borderline"},"baselineRevenueRank":null},{"entityId":"transperfect-dataforce","name":"TransPerfect DataForce","reachPct":1.059066,"importance":0.290662171658,"importanceRank":18,"functions":[{"id":"expert-data","weight":0.629,"sharePct":0.339,"shareLowPct":0.071,"shareHighPct":1.259,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"labelling-multimodal","weight":0.155,"sharePct":5.457,"shareLowPct":2.061,"shareHighPct":13.966,"stale":false,"offBasis":false,"level":1.16666666667,"years":0.2806155120779916,"disagreement":false},{"id":"rl-environments","weight":0.147,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"evaluation","weight":0.045,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"data-tooling","weight":0.024,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false}],"rising":null,"risingEligible":false,"risingComponents":{},"mc":{"medianRank":18,"rankP10":13,"rankP90":22,"pTop8":0.012,"pSelected":0.0156,"stability":"borderline"},"baselineRevenueRank":null},{"entityId":"toloka","name":"Toloka","reachPct":0.962138,"importance":0.2405345,"importanceRank":19,"functions":[{"id":"expert-data","weight":0.629,"sharePct":1.087,"shareLowPct":0.0956,"shareHighPct":3.1604,"stale":true,"offBasis":false,"level":1,"years":0.25,"disagreement":false},{"id":"labelling-multimodal","weight":0.155,"sharePct":1.251,"shareLowPct":0,"shareHighPct":4.3882,"stale":true,"offBasis":false,"level":1,"years":0.25,"disagreement":false},{"id":"rl-environments","weight":0.147,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"evaluation","weight":0.045,"sharePct":1.878,"shareLowPct":0,"shareHighPct":6.4996,"stale":true,"offBasis":false,"level":1,"years":0.25,"disagreement":false},{"id":"data-tooling","weight":0.024,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false}],"rising":null,"risingEligible":false,"risingComponents":{},"mc":{"medianRank":20,"rankP10":16,"rankP90":25,"pTop8":0.0002,"pSelected":0.0002,"stability":"borderline"},"baselineRevenueRank":null},{"entityId":"fleet-ai","name":"Fleet","reachPct":0.799386,"importance":0.224320111738,"importanceRank":20,"functions":[{"id":"expert-data","weight":0.629,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"labelling-multimodal","weight":0.155,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"rl-environments","weight":0.147,"sharePct":5.438,"shareLowPct":3.335,"shareHighPct":10.315,"stale":false,"offBasis":false,"level":1.16666666667,"years":0.2806155120779916,"disagreement":false},{"id":"evaluation","weight":0.045,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"data-tooling","weight":0.024,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false}],"rising":null,"risingEligible":false,"risingComponents":{},"mc":{"medianRank":22,"rankP10":18,"rankP90":27,"pTop8":0.0001,"pSelected":0.0008,"stability":"borderline"},"baselineRevenueRank":15},{"entityId":"uber-ai-solutions","name":"Uber AI Solutions","reachPct":0.706596,"importance":0.176649,"importanceRank":21,"functions":[{"id":"expert-data","weight":0.629,"sharePct":0.339,"shareLowPct":0.042,"shareHighPct":1.568,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"labelling-multimodal","weight":0.155,"sharePct":3.183,"shareLowPct":0.642,"shareHighPct":11.092,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false},{"id":"rl-environments","weight":0.147,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"evaluation","weight":0.045,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"data-tooling","weight":0.024,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false}],"rising":null,"risingEligible":false,"risingComponents":{},"mc":{"medianRank":21,"rankP10":17,"rankP90":26,"pTop8":0.0001,"pSelected":0.0002,"stability":"borderline"},"baselineRevenueRank":null},{"entityId":"welo-data","name":"Welo Data","reachPct":0.695566,"importance":0.1738915,"importanceRank":22,"functions":[{"id":"expert-data","weight":0.629,"sharePct":0.339,"shareLowPct":0.064,"shareHighPct":1.383,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"labelling-multimodal","weight":0.155,"sharePct":2.501,"shareLowPct":0.565,"shareHighPct":9.041,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false},{"id":"rl-environments","weight":0.147,"sharePct":0.435,"shareLowPct":0.082,"shareHighPct":1.771,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"evaluation","weight":0.045,"sharePct":0.683,"shareLowPct":0.129,"shareHighPct":2.757,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"data-tooling","weight":0.024,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false}],"rising":null,"risingEligible":false,"risingComponents":{},"mc":{"medianRank":21,"rankP10":17,"rankP90":25,"pTop8":0,"pSelected":0.0001,"stability":"borderline"},"baselineRevenueRank":null},{"entityId":"imerit","name":"iMerit","reachPct":0.639358,"importance":0.1598395,"importanceRank":23,"functions":[{"id":"expert-data","weight":0.629,"sharePct":0.356,"shareLowPct":0.124,"shareHighPct":0.811,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"labelling-multimodal","weight":0.155,"sharePct":2.046,"shareLowPct":0.72,"shareHighPct":4.569,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false},{"id":"rl-environments","weight":0.147,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"evaluation","weight":0.045,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"data-tooling","weight":0.024,"sharePct":4.096,"shareLowPct":1.461,"shareHighPct":8.916,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false}],"rising":null,"risingEligible":false,"risingComponents":{},"mc":{"medianRank":25,"rankP10":22,"rankP90":28,"pTop8":0,"pSelected":0,"stability":"borderline"},"baselineRevenueRank":null},{"entityId":"lionbridge","name":"Lionbridge (AI training services)","reachPct":0.613903,"importance":0.15347575,"importanceRank":24,"functions":[{"id":"expert-data","weight":0.629,"sharePct":0.247,"shareLowPct":0.035,"shareHighPct":1.05,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"labelling-multimodal","weight":0.155,"sharePct":2.785,"shareLowPct":0.639,"shareHighPct":9.041,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false},{"id":"rl-environments","weight":0.147,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"evaluation","weight":0.045,"sharePct":0.597,"shareLowPct":0.086,"shareHighPct":2.511,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"data-tooling","weight":0.024,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false}],"rising":null,"risingEligible":false,"risingComponents":{},"mc":{"medianRank":22,"rankP10":19,"rankP90":28,"pTop8":0,"pSelected":0,"stability":"borderline"},"baselineRevenueRank":null},{"entityId":"lxt","name":"LXT","reachPct":0.609992,"importance":0.152498,"importanceRank":25,"functions":[{"id":"expert-data","weight":0.629,"sharePct":0.198,"shareLowPct":0.035,"shareHighPct":0.758,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"labelling-multimodal","weight":0.155,"sharePct":2.785,"shareLowPct":0.798,"shareHighPct":8.211,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false},{"id":"rl-environments","weight":0.147,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"evaluation","weight":0.045,"sharePct":1.195,"shareLowPct":0.215,"shareHighPct":4.456,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false},{"id":"data-tooling","weight":0.024,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false}],"rising":null,"risingEligible":false,"risingComponents":{},"mc":{"medianRank":23,"rankP10":20,"rankP90":28,"pTop8":0,"pSelected":0,"stability":"borderline"},"baselineRevenueRank":null},{"entityId":"dataocean-ai","name":"DataOcean AI","reachPct":0.475882,"importance":0.132808099149,"importanceRank":26,"functions":[{"id":"expert-data","weight":0.629,"sharePct":0.038,"shareLowPct":0.016,"shareHighPct":0.069,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"labelling-multimodal","weight":0.155,"sharePct":2.916,"shareLowPct":2.379,"shareHighPct":3.578,"stale":false,"offBasis":false,"level":1.16666666667,"years":0.2806155120779916,"disagreement":false},{"id":"rl-environments","weight":0.147,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"evaluation","weight":0.045,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"data-tooling","weight":0.024,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false}],"rising":null,"risingEligible":false,"risingComponents":{},"mc":{"medianRank":29,"rankP10":26,"rankP90":32,"pTop8":0,"pSelected":0,"stability":"borderline"},"baselineRevenueRank":null},{"entityId":"encord","name":"Encord","reachPct":0.362881,"importance":0.100752830846,"importanceRank":27,"functions":[{"id":"expert-data","weight":0.629,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"labelling-multimodal","weight":0.155,"sharePct":0.227,"shareLowPct":0.057,"shareHighPct":0.679,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"rl-environments","weight":0.147,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"evaluation","weight":0.045,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"data-tooling","weight":0.024,"sharePct":13.654,"shareLowPct":6.949,"shareHighPct":25.029,"stale":false,"offBasis":false,"level":1.16666666667,"years":0.2806155120779916,"disagreement":false}],"rising":null,"risingEligible":false,"risingComponents":{},"mc":{"medianRank":30,"rankP10":24,"rankP90":34,"pTop8":0,"pSelected":0,"stability":"borderline"},"baselineRevenueRank":null},{"entityId":"datatang","name":"Datatang","reachPct":0.353055,"importance":0.0974366166512,"importanceRank":28,"functions":[{"id":"expert-data","weight":0.629,"sharePct":0.056,"shareLowPct":0.011,"shareHighPct":0.211,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"labelling-multimodal","weight":0.155,"sharePct":1.933,"shareLowPct":0.669,"shareHighPct":5.089,"stale":false,"offBasis":false,"level":1.16666666667,"years":0.2806155120779916,"disagreement":false},{"id":"rl-environments","weight":0.147,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"evaluation","weight":0.045,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"data-tooling","weight":0.024,"sharePct":0.759,"shareLowPct":0.143,"shareHighPct":2.787,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true}],"rising":null,"risingEligible":false,"risingComponents":{},"mc":{"medianRank":29,"rankP10":24,"rankP90":34,"pTop8":0,"pSelected":0,"stability":"borderline"},"baselineRevenueRank":null},{"entityId":"sama","name":"Sama","reachPct":0.350755,"importance":0.0963206405528,"importanceRank":29,"functions":[{"id":"expert-data","weight":0.629,"sharePct":0.085,"shareLowPct":0.011,"shareHighPct":0.38,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"labelling-multimodal","weight":0.155,"sharePct":1.819,"shareLowPct":0.404,"shareHighPct":5.885,"stale":false,"offBasis":false,"level":1.16666666667,"years":0.2806155120779916,"disagreement":false},{"id":"rl-environments","weight":0.147,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"evaluation","weight":0.045,"sharePct":0.341,"shareLowPct":0.043,"shareHighPct":1.518,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"data-tooling","weight":0.024,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false}],"rising":null,"risingEligible":false,"risingComponents":{},"mc":{"medianRank":28,"rankP10":23,"rankP90":33,"pTop8":0,"pSelected":0,"stability":"borderline"},"baselineRevenueRank":null},{"entityId":"halluminate","name":"Halluminate","reachPct":0.278166,"importance":0.0695415,"importanceRank":30,"functions":[{"id":"expert-data","weight":0.629,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"labelling-multimodal","weight":0.155,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"rl-environments","weight":0.147,"sharePct":1.618,"shareLowPct":0.638,"shareHighPct":4.818,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false},{"id":"evaluation","weight":0.045,"sharePct":0.896,"shareLowPct":0.193,"shareHighPct":3.695,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"data-tooling","weight":0.024,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false}],"rising":null,"risingEligible":false,"risingComponents":{},"mc":{"medianRank":32,"rankP10":26,"rankP90":36,"pTop8":0,"pSelected":0,"stability":"borderline"},"baselineRevenueRank":null},{"entityId":"gray-swan","name":"Gray Swan AI","reachPct":0.15363,"importance":0.0684343850338,"importanceRank":31,"functions":[{"id":"expert-data","weight":0.629,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"labelling-multimodal","weight":0.155,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"rl-environments","weight":0.147,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"evaluation","weight":0.045,"sharePct":3.414,"shareLowPct":0.527,"shareHighPct":9.587,"stale":false,"offBasis":false,"level":1.83333333333,"years":0.4454493590691404,"disagreement":false},{"id":"data-tooling","weight":0.024,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false}],"rising":null,"risingEligible":false,"risingComponents":{},"mc":{"medianRank":34,"rankP10":28,"rankP90":37,"pTop8":0,"pSelected":0,"stability":"borderline"},"baselineRevenueRank":null},{"entityId":"superannotate","name":"SuperAnnotate","reachPct":0.271043,"importance":0.06776075,"importanceRank":32,"functions":[{"id":"expert-data","weight":0.629,"sharePct":0.127,"shareLowPct":0,"shareHighPct":0.4684,"stale":true,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"labelling-multimodal","weight":0.155,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"rl-environments","weight":0.147,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"evaluation","weight":0.045,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"data-tooling","weight":0.024,"sharePct":7.965,"shareLowPct":0.623,"shareHighPct":21.285,"stale":true,"offBasis":false,"level":1,"years":0.25,"disagreement":false}],"rising":null,"risingEligible":false,"risingComponents":{},"mc":{"medianRank":32,"rankP10":27,"rankP90":36,"pTop8":0,"pSelected":0,"stability":"borderline"},"baselineRevenueRank":null},{"entityId":"irregular","name":"Irregular","reachPct":0.09216,"importance":0.0651669609542,"importanceRank":33,"functions":[{"id":"expert-data","weight":0.629,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"labelling-multimodal","weight":0.155,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"rl-environments","weight":0.147,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"evaluation","weight":0.045,"sharePct":2.048,"shareLowPct":0.1104,"shareHighPct":6.9256,"stale":true,"offBasis":false,"level":2.5,"years":0.7071067811865475,"disagreement":false},{"id":"data-tooling","weight":0.024,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false}],"rising":null,"risingEligible":false,"risingComponents":{},"mc":{"medianRank":34,"rankP10":27,"rankP90":37,"pTop8":0,"pSelected":0,"stability":"borderline"},"baselineRevenueRank":null},{"entityId":"cloudfactory","name":"CloudFactory","reachPct":0.22026,"importance":0.055065,"importanceRank":34,"functions":[{"id":"expert-data","weight":0.629,"sharePct":0.035,"shareLowPct":0.004,"shareHighPct":0.169,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"labelling-multimodal","weight":0.155,"sharePct":1.279,"shareLowPct":0.244,"shareHighPct":4.192,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false},{"id":"rl-environments","weight":0.147,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"evaluation","weight":0.045,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"data-tooling","weight":0.024,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false}],"rising":null,"risingEligible":false,"risingComponents":{},"mc":{"medianRank":34,"rankP10":29,"rankP90":37,"pTop8":0,"pSelected":0,"stability":"borderline"},"baselineRevenueRank":null},{"entityId":"afterquery","name":"AfterQuery","reachPct":0.210664,"importance":0.052666,"importanceRank":35,"functions":[{"id":"expert-data","weight":0.629,"sharePct":0.176,"shareLowPct":0.018,"shareHighPct":1.049,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"labelling-multimodal","weight":0.155,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"rl-environments","weight":0.147,"sharePct":0.68,"shareLowPct":0.068,"shareHighPct":3.945,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"evaluation","weight":0.045,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"data-tooling","weight":0.024,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false}],"rising":null,"risingEligible":false,"risingComponents":{},"mc":{"medianRank":29,"rankP10":23,"rankP90":35,"pTop8":0,"pSelected":0,"stability":"borderline"},"baselineRevenueRank":null},{"entityId":"artificial-analysis","name":"Artificial Analysis","reachPct":0.1152,"importance":0.0513157661648,"importanceRank":36,"functions":[{"id":"expert-data","weight":0.629,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"labelling-multimodal","weight":0.155,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"rl-environments","weight":0.147,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"evaluation","weight":0.045,"sharePct":2.56,"shareLowPct":0.523,"shareHighPct":8.053,"stale":false,"offBasis":false,"level":1.83333333333,"years":0.4454493590691404,"disagreement":false},{"id":"data-tooling","weight":0.024,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false}],"rising":null,"risingEligible":false,"risingComponents":{},"mc":{"medianRank":35,"rankP10":29,"rankP90":37,"pTop8":0,"pSelected":0,"stability":"borderline"},"baselineRevenueRank":null},{"entityId":"datacurve","name":"Datacurve","reachPct":0.167789,"importance":0.04194725,"importanceRank":37,"functions":[{"id":"expert-data","weight":0.629,"sharePct":0.127,"shareLowPct":0.01,"shareHighPct":0.757,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"labelling-multimodal","weight":0.155,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"rl-environments","weight":0.147,"sharePct":0.598,"shareLowPct":0.053,"shareHighPct":3.281,"stale":false,"offBasis":false,"level":1,"years":0.25,"disagreement":false,"defaultLevel":true},{"id":"evaluation","weight":0.045,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false},{"id":"data-tooling","weight":0.024,"sharePct":0,"shareLowPct":0,"shareHighPct":0,"stale":false,"offBasis":false,"level":null,"years":null,"disagreement":false}],"rising":null,"risingEligible":false,"risingComponents":{},"mc":{"medianRank":32,"rankP10":26,"rankP90":36,"pTop8":0,"pSelected":0,"stability":"borderline"},"baselineRevenueRank":null}],"selection":{"importanceIds":["mercor","surge-ai","handshake-ai","micro1","scale-ai","arena-ai","snorkel-ai","turing"],"risingIds":["proximal"],"filledFromImportance":["innodata"]},"risingPools":{"growth":12,"shareChange":8,"committed":0},"weightSensitivity":[{"functionId":"expert-data","end":"low","entered":["appen"],"left":["turing"]},{"functionId":"expert-data","end":"high","entered":["prolific"],"left":["arena-ai"]},{"functionId":"labelling-multimodal","end":"low","entered":[],"left":[]},{"functionId":"labelling-multimodal","end":"high","entered":["appen","taskus"],"left":["arena-ai","turing"]},{"functionId":"rl-environments","end":"low","entered":["prolific"],"left":["turing"]},{"functionId":"rl-environments","end":"high","entered":[],"left":[]},{"functionId":"evaluation","end":"low","entered":["prolific"],"left":["arena-ai"]},{"functionId":"evaluation","end":"high","entered":[],"left":[]},{"functionId":"data-tooling","end":"low","entered":[],"left":[]},{"functionId":"data-tooling","end":"high","entered":[],"left":[]}],"validation":[],"baseline":{"spearman":0.842857142857,"baselineTop10NotSelected":["prolific","taskus"],"selectedNotInBaselineTop10":["arena-ai","proximal"]},"raterAgreement":{"within":0.940298507463,"weightedKappa":0.705586942875,"disagreements":[]}},"inputs":{"mercor":{"schema":"5lc-index-inputs/1","key":"models-data-evaluation","entityId":"mercor","name":"Mercor","kind":"private","commercialActivityVerified":true,"shares":{"expert-data":{"low":18.7,"central":23.291,"high":31.074,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"Annualised gross run-rate, June 2026, extended one quarter (estimate)","periodEnd":"2026-06-30","source":{"url":"https://techcrunch.com/2026/07/08/these-ai-startups-are-growing-revenue-at-faster-and-faster-rates/","title":"These AI startups are growing revenue at faster and faster rates","publisher":"TechCrunch","publishedDate":"2026-07-08","accessedDate":"2026-10-22","tier":"T5","locator":"The company says (CEO Brendan Foody, post on X): Mercor crossed $2 billion in gross annualized revenue as of June 2026, about four months after $1 billion; \"$500 million run rate in September\" (2025)"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Method: The company says (CEO Brendan Foody, TechCrunch, 8 Jul 2026) its gross annualised revenue crossed US$2bn in June 2026, four months after US$1bn; Dealroom (6 Jul 2026, reporting The Information, T3) reports 'over US$2bn' on the same gross basis. Central US$2.2bn extends the stated June figure by one quarter of continued growth (an estimate), low is the stated US$2.0bn, high US$2.8bn. Contractors take 60-70% of this gross figure (not deducted; the card is gross). Sepal AI and Deeptune are counted here (consolidated). Gross revenue of about US$614M for H1 2026 (Sacra, T4) is consistent with a run-rate that doubled during the half. Split follows the card notes (75/22/3). Full pass: no newer disclosure than June 2026 was found. Main input: TechCrunch (T2). Share = company ÷ function total; low and high move only this company's own range and hold the others at central. Mercor expert data for post-training: US$1,650m (1,250-2,450m) = run-rate US$2,200m (2,000-2,800m) × 75% (the smaller parts of a mixed vendor are varied by ±50% and the largest part absorbs the difference). Function total, annualised US$m (central): 7,084, made of identified candidates 6,934 and an allowance of 150 for vendors outside the register (for example Deccan AI, Shaip, Cogito Tech, Sigma AI, Centaur Labs, Revelo, e2f, Patronus AI, Haize Labs, Lightly, Mostly AI, Abundant, Ethos, Datumo and smaller environment builders). Largest, US$m: mercor 1,650, surge-ai 1,296, handshake-ai 882, micro1 850, scale-ai 412, other candidates 1,844, vendors outside the register 150. Full table: a local evidence capture. Share 23.291% (range 18.7-31.074%). Source published date: 2026-07-08 (full date)."},"labelling-multimodal":{"participates":false,"evidence":{"url":"https://www.mercor.com/","title":"Mercor homepage","publisher":"Mercor","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T2","locator":"Product list: expert evaluation and data, APEX benchmarks and RL environments, enterprise AI, data licensing; no image/video/speech labelling or data software; no large-scale labelling and multimodal data offering described"}},"rl-environments":{"low":13.97,"central":26.321,"high":40.547,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"Annualised gross run-rate, June 2026, extended one quarter (estimate)","periodEnd":"2026-06-30","source":{"url":"https://techcrunch.com/2026/07/08/these-ai-startups-are-growing-revenue-at-faster-and-faster-rates/","title":"These AI startups are growing revenue at faster and faster rates","publisher":"TechCrunch","publishedDate":"2026-07-08","accessedDate":"2026-10-22","tier":"T5","locator":"The company says (CEO Brendan Foody, post on X): Mercor crossed $2 billion in gross annualized revenue as of June 2026, about four months after $1 billion; \"$500 million run rate in September\" (2025)"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Method: The company says (CEO Brendan Foody, TechCrunch, 8 Jul 2026) its gross annualised revenue crossed US$2bn in June 2026, four months after US$1bn; Dealroom (6 Jul 2026, reporting The Information, T3) reports 'over US$2bn' on the same gross basis. Central US$2.2bn extends the stated June figure by one quarter of continued growth (an estimate), low is the stated US$2.0bn, high US$2.8bn. Contractors take 60-70% of this gross figure (not deducted; the card is gross). Sepal AI and Deeptune are counted here (consolidated). Gross revenue of about US$614M for H1 2026 (Sacra, T4) is consistent with a run-rate that doubled during the half. Split follows the card notes (75/22/3). Full pass: no newer disclosure than June 2026 was found. Main input: TechCrunch (T2). Share = company ÷ function total; low and high move only this company's own range and hold the others at central. Mercor RL environments and agent tasks: US$484m (220-924m) = run-rate US$2,200m (2,000-2,800m) × 22% (the smaller parts of a mixed vendor are varied by ±50% and the largest part absorbs the difference). Function total, annualised US$m (central): 1,839, made of identified candidates 1,739 and an allowance of 100 for vendors outside the register (for example Deccan AI, Shaip, Cogito Tech, Sigma AI, Centaur Labs, Revelo, e2f, Patronus AI, Haize Labs, Lightly, Mostly AI, Abundant, Ethos, Datumo and smaller environment builders). Largest, US$m: mercor 484, surge-ai 414, proximal 140, handshake-ai 136, micro1 130, other candidates 434, vendors outside the register 100. Full table: a local evidence capture. Share 26.321% (range 13.97-40.547%). Source published date: 2026-07-08 (full date)."},"evaluation":{"low":5.456,"central":11.265,"high":19.508,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"Annualised gross run-rate, June 2026, extended one quarter (estimate)","periodEnd":"2026-06-30","source":{"url":"https://techcrunch.com/2026/07/08/these-ai-startups-are-growing-revenue-at-faster-and-faster-rates/","title":"These AI startups are growing revenue at faster and faster rates","publisher":"TechCrunch","publishedDate":"2026-07-08","accessedDate":"2026-10-22","tier":"T5","locator":"The company says (CEO Brendan Foody, post on X): Mercor crossed $2 billion in gross annualized revenue as of June 2026, about four months after $1 billion; \"$500 million run rate in September\" (2025)"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Method: The company says (CEO Brendan Foody, TechCrunch, 8 Jul 2026) its gross annualised revenue crossed US$2bn in June 2026, four months after US$1bn; Dealroom (6 Jul 2026, reporting The Information, T3) reports 'over US$2bn' on the same gross basis. Central US$2.2bn extends the stated June figure by one quarter of continued growth (an estimate), low is the stated US$2.0bn, high US$2.8bn. Contractors take 60-70% of this gross figure (not deducted; the card is gross). Sepal AI and Deeptune are counted here (consolidated). Gross revenue of about US$614M for H1 2026 (Sacra, T4) is consistent with a run-rate that doubled during the half. Split follows the card notes (75/22/3). Full pass: no newer disclosure than June 2026 was found. Main input: TechCrunch (T2). Share = company ÷ function total; low and high move only this company's own range and hold the others at central. Mercor independent model evaluation: US$66m (30-126m) = run-rate US$2,200m (2,000-2,800m) × 3% (the smaller parts of a mixed vendor are varied by ±50% and the largest part absorbs the difference). Function total, annualised US$m (central): 586, made of identified candidates 566 and an allowance of 20 for vendors outside the register (for example Deccan AI, Shaip, Cogito Tech, Sigma AI, Centaur Labs, Revelo, e2f, Patronus AI, Haize Labs, Lightly, Mostly AI, Abundant, Ethos, Datumo and smaller environment builders). Largest, US$m: arena-ai 110, surge-ai 90, mercor 66, scale-ai 45, innodata 33, other candidates 222, vendors outside the register 20. Full table: a local evidence capture. Share 11.265% (range 5.456-19.508%). Source published date: 2026-07-08 (full date)."},"data-tooling":{"participates":false,"evidence":{"url":"https://www.mercor.com/","title":"Mercor homepage","publisher":"Mercor","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T2","locator":"Product list: expert evaluation and data, APEX benchmarks and RL environments, enterprise AI, data licensing; no image/video/speech labelling or data software; no labelling and data software offering described"}}},"priorShares":{"expert-data":{"low":9.734,"central":12.571,"high":14.365,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"September 2025 run-rate (Q3 2025 frame)","periodEnd":"2025-09-30","source":{"url":"https://techcrunch.com/2026/07/08/these-ai-startups-are-growing-revenue-at-faster-and-faster-rates/","title":"These AI startups are growing revenue at faster and faster rates","publisher":"TechCrunch","publishedDate":"2026-07-08","accessedDate":"2026-10-22","tier":"T5","locator":"The company says its run-rate was \"$500 million in September\" 2025 (CEO Foody); TechCrunch of 9 Sep 2025 had \"approaching $450 million\" from one source"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Prior share in the Q3 2025 frame. The company's September 2025 run-rate was US$500M (the company says, via TechCrunch, 8 Jul 2026, T2); TechCrunch (9 Sep 2025, T3) had 'approaching US$450M' from one source, which sets the low end (US$450-500M). Mercor expert data for post-training: US$375m (281-438m) = prior run-rate US$500m (450-500m) × 75% (split and the ±50% variation of the smaller parts held as today). Prior part total US$2,983m = the sum over all vendors of Q3 2025 run-rate × split, plus the allowance of 150 ÷ 1.5 = 100. Q3 2025 frame: each vendor's run-rate at the time, documented where it exists (Mercor US$500M, Handshake AI US$80M, micro1 US$45M, Snorkel US$20.8M, Arena and Proximal none, Appen, TaskUs and Innodata from their H1 or Q2 2025 reports scaled to today's basis, Surge US$1.2bn, Scale's data business US$850M, Turing US$337.5M, Fleet US$1M, Labelbox one fifth of today's, RWS today's divided by 1.45), and every other vendor at today's run-rate divided by 1.5 (the rounded median, 1.54, of the 12 documented multiples); each vendor's split across the five parts is held at today's. The frame is an estimate and its uncertainty is not in the share ranges. Share 12.571% (range 9.734-14.365%). The current share is 23.291% on the same basis, so the change is +10.7 points."},"rl-environments":{"low":7.668,"central":15.579,"high":21.68,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"September 2025 run-rate (Q3 2025 frame)","periodEnd":"2025-09-30","source":{"url":"https://techcrunch.com/2026/07/08/these-ai-startups-are-growing-revenue-at-faster-and-faster-rates/","title":"These AI startups are growing revenue at faster and faster rates","publisher":"TechCrunch","publishedDate":"2026-07-08","accessedDate":"2026-10-22","tier":"T5","locator":"The company says its run-rate was \"$500 million in September\" 2025 (CEO Foody); TechCrunch of 9 Sep 2025 had \"approaching $450 million\" from one source"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Prior share in the Q3 2025 frame. The company's September 2025 run-rate was US$500M (the company says, via TechCrunch, 8 Jul 2026, T2); TechCrunch (9 Sep 2025, T3) had 'approaching US$450M' from one source, which sets the low end (US$450-500M). Mercor RL environments and agent tasks: US$110m (49.5-165m) = prior run-rate US$500m (450-500m) × 22% (split and the ±50% variation of the smaller parts held as today). Prior part total US$706m = the sum over all vendors of Q3 2025 run-rate × split, plus the allowance of 100 ÷ 1.5 = 66.7. Q3 2025 frame: each vendor's run-rate at the time, documented where it exists (Mercor US$500M, Handshake AI US$80M, micro1 US$45M, Snorkel US$20.8M, Arena and Proximal none, Appen, TaskUs and Innodata from their H1 or Q2 2025 reports scaled to today's basis, Surge US$1.2bn, Scale's data business US$850M, Turing US$337.5M, Fleet US$1M, Labelbox one fifth of today's, RWS today's divided by 1.45), and every other vendor at today's run-rate divided by 1.5 (the rounded median, 1.54, of the 12 documented multiples); each vendor's split across the five parts is held at today's. The frame is an estimate and its uncertainty is not in the share ranges. Share 15.579% (range 7.668-21.68%). The current share is 26.321% on the same basis, so the change is +10.7 points."},"evaluation":{"low":2.629,"central":5.66,"high":8.256,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"September 2025 run-rate (Q3 2025 frame)","periodEnd":"2025-09-30","source":{"url":"https://techcrunch.com/2026/07/08/these-ai-startups-are-growing-revenue-at-faster-and-faster-rates/","title":"These AI startups are growing revenue at faster and faster rates","publisher":"TechCrunch","publishedDate":"2026-07-08","accessedDate":"2026-10-22","tier":"T5","locator":"The company says its run-rate was \"$500 million in September\" 2025 (CEO Foody); TechCrunch of 9 Sep 2025 had \"approaching $450 million\" from one source"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Prior share in the Q3 2025 frame. The company's September 2025 run-rate was US$500M (the company says, via TechCrunch, 8 Jul 2026, T2); TechCrunch (9 Sep 2025, T3) had 'approaching US$450M' from one source, which sets the low end (US$450-500M). Mercor independent model evaluation: US$15m (6.75-22.5m) = prior run-rate US$500m (450-500m) × 3% (split and the ±50% variation of the smaller parts held as today). Prior part total US$265m = the sum over all vendors of Q3 2025 run-rate × split, plus the allowance of 20 ÷ 1.5 = 13.3. Q3 2025 frame: each vendor's run-rate at the time, documented where it exists (Mercor US$500M, Handshake AI US$80M, micro1 US$45M, Snorkel US$20.8M, Arena and Proximal none, Appen, TaskUs and Innodata from their H1 or Q2 2025 reports scaled to today's basis, Surge US$1.2bn, Scale's data business US$850M, Turing US$337.5M, Fleet US$1M, Labelbox one fifth of today's, RWS today's divided by 1.45), and every other vendor at today's run-rate divided by 1.5 (the rounded median, 1.54, of the 12 documented multiples); each vendor's split across the five parts is held at today's. The frame is an estimate and its uncertainty is not in the share ranges. Share 5.66% (range 2.629-8.256%). The current share is 11.265% on the same basis, so the change is +5.6 points."}},"activityGrowth":{"current":{"low":2000000000,"central":2000000000,"high":2000000000,"unit":"USD","basis":"Gross annualised revenue run-rate in US dollars, before payments to contractors (latest month × 12 or latest quarter × 4)","period":"June 2026 gross run-rate","periodEnd":"2026-06-30","source":{"url":"https://techcrunch.com/2026/07/08/these-ai-startups-are-growing-revenue-at-faster-and-faster-rates/","title":"These AI startups are growing revenue at faster and faster rates","publisher":"TechCrunch","publishedDate":"2026-07-08","accessedDate":"2026-10-22","tier":"T2","locator":"The company says (CEO Brendan Foody, post on X): Mercor crossed $2 billion in gross annualized revenue as of June 2026, about four months after $1 billion; \"$500 million run rate in September\" (2025)"},"note":"The company says (CEO Foody, via TechCrunch, 8 Jul 2026) gross annualised revenue crossed US$2bn in June 2026."},"prior":{"low":450000000,"central":500000000,"high":500000000,"unit":"USD","basis":"Gross annualised revenue run-rate in US dollars, before payments to contractors (latest month × 12 or latest quarter × 4)","period":"September 2025 gross run-rate","periodEnd":"2025-09-30","source":{"url":"https://techcrunch.com/2026/07/08/these-ai-startups-are-growing-revenue-at-faster-and-faster-rates/","title":"These AI startups are growing revenue at faster and faster rates","publisher":"TechCrunch","publishedDate":"2026-07-08","accessedDate":"2026-10-22","tier":"T2","locator":"The company says (CEO Brendan Foody, post on X): Mercor crossed $2 billion in gross annualized revenue as of June 2026, about four months after $1 billion; \"$500 million run rate in September\" (2025)"},"note":"The same TechCrunch article gives the company's September 2025 run-rate as US$500M; TechCrunch (9 Sep 2025, T3) had \"approaching US$450M\" from one source while the CEO said the figure was higher, which sets the low end. Growth is 4.0 times, equal to the engine's +300% cap."}},"baselineRevenueUsd":{"low":2000000000,"central":2000000000,"high":2000000000,"unit":"USD","basis":"Gross annualised revenue run-rate in US dollars, before payments to contractors (latest month × 12 or latest quarter × 4)","period":"June 2026 gross run-rate","periodEnd":"2026-06-30","source":{"url":"https://techcrunch.com/2026/07/08/these-ai-startups-are-growing-revenue-at-faster-and-faster-rates/","title":"These AI startups are growing revenue at faster and faster rates","publisher":"TechCrunch","publishedDate":"2026-07-08","accessedDate":"2026-10-22","tier":"T2","locator":"The company says (CEO Brendan Foody, post on X): Mercor crossed $2 billion in gross annualized revenue as of June 2026, about four months after $1 billion; \"$500 million run rate in September\" (2025)"},"note":"The company says gross annualised revenue crossed US$2bn in June 2026; annual revenue is not published, so the stated run-rate is used. Gross before contractor pay."},"screening":{"expert-data":{"levelLow":1,"levelHigh":3},"rl-environments":{"levelLow":1,"levelHigh":2},"evaluation":{"levelLow":1,"levelHigh":2}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"766c936d28e20342faae368b40daae790ddf3463fc91dd49b4756e7809e969b5"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"766c936d28e20342faae368b40daae790ddf3463fc91dd49b4756e7809e969b5"}}},"surge-ai":{"schema":"5lc-index-inputs/1","key":"models-data-evaluation","entityId":"surge-ai","name":"Surge AI","kind":"private","commercialActivityVerified":true,"shares":{"expert-data":{"low":9.108,"central":18.294,"high":30.83,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"CY2024 revenue (T3), extrapolated to a 2026 run-rate (estimate)","periodEnd":"2024-12-31","source":{"url":"https://siliconangle.com/2025/07/01/data-labeling-startup-surge-ai-seeking-1b-first-capital-raise-reports-say/","title":"Data labeling startup Surge AI seeking $1B in first capital raise, reports say","publisher":"SiliconANGLE (reporting Reuters)","publishedDate":"2025-07-01","accessedDate":"2026-10-22","tier":"T5","locator":"Surge has already grown its annual revenue to more than $1 billion (Reuters, anonymous sources); Scale generated $870 million in sales the year before"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Method: No current figure is published; the full pass found nothing newer. Evidence: more than US$1bn of annual revenue in July 2025 and profitable (SiliconANGLE, 1 Jul 2025, T3, citing Reuters' anonymous sources; opened); Sacra US$1.2bn for 2024 (T4; opened); SemiAnalysis, 6 Jan 2026: 'closer to $1B ARR', the authors' own estimate (T4; opened). A US$1.4bn figure for August 2025 appears only on aggregator pages and is not used. The central US$1.8bn extrapolates 2024 with the growth of peers over 2025-26 (Mercor US$0.5bn in September 2025 to US$2bn in June 2026, Handshake US$0.08bn in August 2025 to US$1bn), low US$1.0bn is the 2024 level, high US$3.0bn is the card's upper bound (the card notes record an October 2026 interview citing about US$3bn, which the company has not confirmed). Split follows the card notes. Main input: SiliconANGLE (reporting Reuters) (T3). Share = company ÷ function total; low and high move only this company's own range and hold the others at central. Surge AI expert data for post-training: US$1,296m (580-2,580m) = run-rate US$1,800m (1,000-3,000m) × 72% (the smaller parts of a mixed vendor are varied by ±50% and the largest part absorbs the difference). Function total, annualised US$m (central): 7,084, made of identified candidates 6,934 and an allowance of 150 for vendors outside the register (for example Deccan AI, Shaip, Cogito Tech, Sigma AI, Centaur Labs, Revelo, e2f, Patronus AI, Haize Labs, Lightly, Mostly AI, Abundant, Ethos, Datumo and smaller environment builders). Largest, US$m: mercor 1,650, surge-ai 1,296, handshake-ai 882, micro1 850, scale-ai 412, other candidates 1,844, vendors outside the register 150. Full table: a local evidence capture. Share 18.294% (range 9.108-30.83%). Source published date: 2025-07-01 (full date)."},"labelling-multimodal":{"participates":false,"evidence":{"url":"https://surgehq.ai/","title":"Surge AI homepage","publisher":"Surge AI","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T2","locator":"Offerings: off-the-shelf frontier data and RL environments, expert workforce, post-training runs, benchmarks; no revenue or labelling product listed; no large-scale labelling and multimodal data offering described"}},"rl-environments":{"low":7.468,"central":22.514,"high":42.076,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"CY2024 revenue (T3), extrapolated to a 2026 run-rate (estimate)","periodEnd":"2024-12-31","source":{"url":"https://siliconangle.com/2025/07/01/data-labeling-startup-surge-ai-seeking-1b-first-capital-raise-reports-say/","title":"Data labeling startup Surge AI seeking $1B in first capital raise, reports say","publisher":"SiliconANGLE (reporting Reuters)","publishedDate":"2025-07-01","accessedDate":"2026-10-22","tier":"T5","locator":"Surge has already grown its annual revenue to more than $1 billion (Reuters, anonymous sources); Scale generated $870 million in sales the year before"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Method: No current figure is published; the full pass found nothing newer. Evidence: more than US$1bn of annual revenue in July 2025 and profitable (SiliconANGLE, 1 Jul 2025, T3, citing Reuters' anonymous sources; opened); Sacra US$1.2bn for 2024 (T4; opened); SemiAnalysis, 6 Jan 2026: 'closer to $1B ARR', the authors' own estimate (T4; opened). A US$1.4bn figure for August 2025 appears only on aggregator pages and is not used. The central US$1.8bn extrapolates 2024 with the growth of peers over 2025-26 (Mercor US$0.5bn in September 2025 to US$2bn in June 2026, Handshake US$0.08bn in August 2025 to US$1bn), low US$1.0bn is the 2024 level, high US$3.0bn is the card's upper bound (the card notes record an October 2026 interview citing about US$3bn, which the company has not confirmed). Split follows the card notes. Main input: SiliconANGLE (reporting Reuters) (T3). Share = company ÷ function total; low and high move only this company's own range and hold the others at central. Surge AI RL environments and agent tasks: US$414m (115-1,035m) = run-rate US$1,800m (1,000-3,000m) × 23% (the smaller parts of a mixed vendor are varied by ±50% and the largest part absorbs the difference). Function total, annualised US$m (central): 1,839, made of identified candidates 1,739 and an allowance of 100 for vendors outside the register (for example Deccan AI, Shaip, Cogito Tech, Sigma AI, Centaur Labs, Revelo, e2f, Patronus AI, Haize Labs, Lightly, Mostly AI, Abundant, Ethos, Datumo and smaller environment builders). Largest, US$m: mercor 484, surge-ai 414, proximal 140, handshake-ai 136, micro1 130, other candidates 434, vendors outside the register 100. Full table: a local evidence capture. Share 22.514% (range 7.468-42.076%). Source published date: 2025-07-01 (full date)."},"evaluation":{"low":4.8,"central":15.362,"high":31.212,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"CY2024 revenue (T3), extrapolated to a 2026 run-rate (estimate)","periodEnd":"2024-12-31","source":{"url":"https://siliconangle.com/2025/07/01/data-labeling-startup-surge-ai-seeking-1b-first-capital-raise-reports-say/","title":"Data labeling startup Surge AI seeking $1B in first capital raise, reports say","publisher":"SiliconANGLE (reporting Reuters)","publishedDate":"2025-07-01","accessedDate":"2026-10-22","tier":"T5","locator":"Surge has already grown its annual revenue to more than $1 billion (Reuters, anonymous sources); Scale generated $870 million in sales the year before"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Method: No current figure is published; the full pass found nothing newer. Evidence: more than US$1bn of annual revenue in July 2025 and profitable (SiliconANGLE, 1 Jul 2025, T3, citing Reuters' anonymous sources; opened); Sacra US$1.2bn for 2024 (T4; opened); SemiAnalysis, 6 Jan 2026: 'closer to $1B ARR', the authors' own estimate (T4; opened). A US$1.4bn figure for August 2025 appears only on aggregator pages and is not used. The central US$1.8bn extrapolates 2024 with the growth of peers over 2025-26 (Mercor US$0.5bn in September 2025 to US$2bn in June 2026, Handshake US$0.08bn in August 2025 to US$1bn), low US$1.0bn is the 2024 level, high US$3.0bn is the card's upper bound (the card notes record an October 2026 interview citing about US$3bn, which the company has not confirmed). Split follows the card notes. Main input: SiliconANGLE (reporting Reuters) (T3). Share = company ÷ function total; low and high move only this company's own range and hold the others at central. Surge AI independent model evaluation: US$90m (25-225m) = run-rate US$1,800m (1,000-3,000m) × 5% (the smaller parts of a mixed vendor are varied by ±50% and the largest part absorbs the difference). Function total, annualised US$m (central): 586, made of identified candidates 566 and an allowance of 20 for vendors outside the register (for example Deccan AI, Shaip, Cogito Tech, Sigma AI, Centaur Labs, Revelo, e2f, Patronus AI, Haize Labs, Lightly, Mostly AI, Abundant, Ethos, Datumo and smaller environment builders). Largest, US$m: arena-ai 110, surge-ai 90, mercor 66, scale-ai 45, innodata 33, other candidates 222, vendors outside the register 20. Full table: a local evidence capture. Share 15.362% (range 4.8-31.212%). Source published date: 2025-07-01 (full date)."},"data-tooling":{"participates":false,"evidence":{"url":"https://surgehq.ai/","title":"Surge AI homepage","publisher":"Surge AI","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T2","locator":"Offerings: off-the-shelf frontier data and RL environments, expert workforce, post-training runs, benchmarks; no revenue or labelling product listed; no labelling and data software offering described"}}},"baselineRevenueUsd":{"low":1000000000,"central":1000000000,"high":1200000000,"unit":"USD","basis":"Annual revenue, calendar year, US dollars (total company revenue)","period":"CY2024","periodEnd":"2024-12-31","source":{"url":"https://siliconangle.com/2025/07/01/data-labeling-startup-surge-ai-seeking-1b-first-capital-raise-reports-say/","title":"Data labeling startup Surge AI seeking $1B in first capital raise, reports say","publisher":"SiliconANGLE (reporting Reuters)","publishedDate":"2025-07-01","accessedDate":"2026-10-22","tier":"T3","locator":"Surge has already grown its annual revenue to more than $1 billion (Reuters, anonymous sources); Scale generated $870 million in sales the year before"},"note":"Reuters, via SiliconANGLE (T3): more than US$1bn; the high end is Sacra's US$1.2bn estimate for 2024 (T4). No 2025 or 2026 figure is published."},"screening":{"expert-data":{"levelLow":1,"levelHigh":3},"rl-environments":{"levelLow":1,"levelHigh":2},"evaluation":{"levelLow":1,"levelHigh":2}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"15546411361cccd918b3823175048bdd661090cfd484de3773ca9a1b8ec07c2e"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"15546411361cccd918b3823175048bdd661090cfd484de3773ca9a1b8ec07c2e"}}},"handshake-ai":{"schema":"5lc-index-inputs/1","key":"models-data-evaluation","entityId":"handshake-ai","name":"Handshake AI","kind":"unit","parentId":"handshake","commercialActivityVerified":true,"shares":{"expert-data":{"low":10.916,"central":12.45,"high":16.166,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"Annualised gross run-rate, August 2026","periodEnd":"2026-08-31","source":{"url":"https://time.com/collection/time100-ai/2026/garrett-lord/","title":"Garrett Lord: The 100 Most Influential People in AI 2026","publisher":"TIME","publishedDate":"2026-08-27","accessedDate":"2026-10-22","tier":"T5","locator":"The company says (CEO Garrett Lord, quoted by TIME): since launching the AI unit it has grown from zero to more than $1 billion in gross annualized revenue in 16 months, working with eight frontier AI labs"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Method: The company says (CEO Garrett Lord, TIME, 27 Aug 2026) the AI unit went from zero to more than US$1bn in gross annualised revenue in 16 months, working with eight frontier labs; The Information via Dealroom (14 Apr 2026, T3): nearly US$1bn, up from US$550M in January 2026, net about US$300M. Central US$1.05bn (the stated 'more than US$1bn'), low US$1.0bn, high US$1.3bn if growth continued after April. Unit figure only (the parent careers network and its recruiting revenue of about US$150M excluded). Cleanlab (acquired Jan 2026) is counted here. Split follows the card notes. Main input: TIME (T2). Share = company ÷ function total; low and high move only this company's own range and hold the others at central. Handshake AI expert data for post-training: US$882m (760-1,196m) = run-rate US$1,050m (1,000-1,300m) × 84% (the smaller parts of a mixed vendor are varied by ±50% and the largest part absorbs the difference). Function total, annualised US$m (central): 7,084, made of identified candidates 6,934 and an allowance of 150 for vendors outside the register (for example Deccan AI, Shaip, Cogito Tech, Sigma AI, Centaur Labs, Revelo, e2f, Patronus AI, Haize Labs, Lightly, Mostly AI, Abundant, Ethos, Datumo and smaller environment builders). Largest, US$m: mercor 1,650, surge-ai 1,296, handshake-ai 882, micro1 850, scale-ai 412, other candidates 1,844, vendors outside the register 150. Full table: a local evidence capture. Share 12.45% (range 10.916-16.166%). Source published date: 2026-08-27 (full date)."},"labelling-multimodal":{"participates":false,"evidence":{"url":"https://joinhandshake.com/ai/","title":"Handshake AI Fellowship","publisher":"Handshake","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T2","locator":"Paid expert-contributor programme (188k+ fellows, $500M+ payouts, undated); no image/video/speech labelling or data software listed; no large-scale labelling and multimodal data offering described"}},"rl-environments":{"low":3.678,"central":7.423,"high":12.961,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"Annualised gross run-rate, August 2026","periodEnd":"2026-08-31","source":{"url":"https://time.com/collection/time100-ai/2026/garrett-lord/","title":"Garrett Lord: The 100 Most Influential People in AI 2026","publisher":"TIME","publishedDate":"2026-08-27","accessedDate":"2026-10-22","tier":"T5","locator":"The company says (CEO Garrett Lord, quoted by TIME): since launching the AI unit it has grown from zero to more than $1 billion in gross annualized revenue in 16 months, working with eight frontier AI labs"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Method: The company says (CEO Garrett Lord, TIME, 27 Aug 2026) the AI unit went from zero to more than US$1bn in gross annualised revenue in 16 months, working with eight frontier labs; The Information via Dealroom (14 Apr 2026, T3): nearly US$1bn, up from US$550M in January 2026, net about US$300M. Central US$1.05bn (the stated 'more than US$1bn'), low US$1.0bn, high US$1.3bn if growth continued after April. Unit figure only (the parent careers network and its recruiting revenue of about US$150M excluded). Cleanlab (acquired Jan 2026) is counted here. Split follows the card notes. Main input: TIME (T2). Share = company ÷ function total; low and high move only this company's own range and hold the others at central. Handshake AI RL environments and agent tasks: US$136m (65-254m) = run-rate US$1,050m (1,000-1,300m) × 13% (the smaller parts of a mixed vendor are varied by ±50% and the largest part absorbs the difference). Function total, annualised US$m (central): 1,839, made of identified candidates 1,739 and an allowance of 100 for vendors outside the register (for example Deccan AI, Shaip, Cogito Tech, Sigma AI, Centaur Labs, Revelo, e2f, Patronus AI, Haize Labs, Lightly, Mostly AI, Abundant, Ethos, Datumo and smaller environment builders). Largest, US$m: mercor 484, surge-ai 414, proximal 140, handshake-ai 136, micro1 130, other candidates 434, vendors outside the register 100. Full table: a local evidence capture. Share 7.423% (range 3.678-12.961%). Source published date: 2026-08-27 (full date)."},"evaluation":{"low":2.634,"central":5.377,"high":9.545,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"Annualised gross run-rate, August 2026","periodEnd":"2026-08-31","source":{"url":"https://time.com/collection/time100-ai/2026/garrett-lord/","title":"Garrett Lord: The 100 Most Influential People in AI 2026","publisher":"TIME","publishedDate":"2026-08-27","accessedDate":"2026-10-22","tier":"T5","locator":"The company says (CEO Garrett Lord, quoted by TIME): since launching the AI unit it has grown from zero to more than $1 billion in gross annualized revenue in 16 months, working with eight frontier AI labs"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Method: The company says (CEO Garrett Lord, TIME, 27 Aug 2026) the AI unit went from zero to more than US$1bn in gross annualised revenue in 16 months, working with eight frontier labs; The Information via Dealroom (14 Apr 2026, T3): nearly US$1bn, up from US$550M in January 2026, net about US$300M. Central US$1.05bn (the stated 'more than US$1bn'), low US$1.0bn, high US$1.3bn if growth continued after April. Unit figure only (the parent careers network and its recruiting revenue of about US$150M excluded). Cleanlab (acquired Jan 2026) is counted here. Split follows the card notes. Main input: TIME (T2). Share = company ÷ function total; low and high move only this company's own range and hold the others at central. Handshake AI independent model evaluation: US$31.5m (15-58.5m) = run-rate US$1,050m (1,000-1,300m) × 3% (the smaller parts of a mixed vendor are varied by ±50% and the largest part absorbs the difference). Function total, annualised US$m (central): 586, made of identified candidates 566 and an allowance of 20 for vendors outside the register (for example Deccan AI, Shaip, Cogito Tech, Sigma AI, Centaur Labs, Revelo, e2f, Patronus AI, Haize Labs, Lightly, Mostly AI, Abundant, Ethos, Datumo and smaller environment builders). Largest, US$m: arena-ai 110, surge-ai 90, mercor 66, scale-ai 45, innodata 33, other candidates 222, vendors outside the register 20. Full table: a local evidence capture. Share 5.377% (range 2.634-9.545%). Source published date: 2026-08-27 (full date)."},"data-tooling":{"participates":false,"evidence":{"url":"https://joinhandshake.com/ai/","title":"Handshake AI Fellowship","publisher":"Handshake","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T2","locator":"Paid expert-contributor programme (188k+ fellows, $500M+ payouts, undated); no image/video/speech labelling or data software listed; no labelling and data software offering described"}}},"priorShares":{"expert-data":{"low":1.236,"central":2.253,"high":3.413,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"August 2025 run-rate (Q3 2025 frame)","periodEnd":"2025-08-31","source":{"url":"https://sacra.com/research/handshake-vs-mercor/","title":"Handshake vs Mercor","publisher":"Sacra","publishedDate":"2025-09-17","accessedDate":"2026-10-22","tier":"T5","locator":"Sacra estimates Handshake hit $280M in annualized revenue as of August 2025: about $200M from the core job board and $80M from Handshake AI"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Prior share in the Q3 2025 frame. Sacra (T4, 17 Sep 2025) estimates Handshake AI at about US$80M annualised in August 2025, with the tier's 40% range (US$48-112M). Handshake AI expert data for post-training: US$67.2m (36.5-103m) = prior run-rate US$80m (48-112m) × 84% (split and the ±50% variation of the smaller parts held as today). Prior part total US$2,983m = the sum over all vendors of Q3 2025 run-rate × split, plus the allowance of 150 ÷ 1.5 = 100. Q3 2025 frame: each vendor's run-rate at the time, documented where it exists (Mercor US$500M, Handshake AI US$80M, micro1 US$45M, Snorkel US$20.8M, Arena and Proximal none, Appen, TaskUs and Innodata from their H1 or Q2 2025 reports scaled to today's basis, Surge US$1.2bn, Scale's data business US$850M, Turing US$337.5M, Fleet US$1M, Labelbox one fifth of today's, RWS today's divided by 1.45), and every other vendor at today's run-rate divided by 1.5 (the rounded median, 1.54, of the 12 documented multiples); each vendor's split across the five parts is held at today's. The frame is an estimate and its uncertainty is not in the share ranges. Share 2.253% (range 1.236-3.413%). The current share is 12.45% on the same basis, so the change is +10.2 points."},"rl-environments":{"low":0.446,"central":1.473,"high":3.044,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"August 2025 run-rate (Q3 2025 frame)","periodEnd":"2025-08-31","source":{"url":"https://sacra.com/research/handshake-vs-mercor/","title":"Handshake vs Mercor","publisher":"Sacra","publishedDate":"2025-09-17","accessedDate":"2026-10-22","tier":"T5","locator":"Sacra estimates Handshake hit $280M in annualized revenue as of August 2025: about $200M from the core job board and $80M from Handshake AI"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Prior share in the Q3 2025 frame. Sacra (T4, 17 Sep 2025) estimates Handshake AI at about US$80M annualised in August 2025, with the tier's 40% range (US$48-112M). Handshake AI RL environments and agent tasks: US$10.4m (3.12-21.8m) = prior run-rate US$80m (48-112m) × 13% (split and the ±50% variation of the smaller parts held as today). Prior part total US$706m = the sum over all vendors of Q3 2025 run-rate × split, plus the allowance of 100 ÷ 1.5 = 66.7. Q3 2025 frame: each vendor's run-rate at the time, documented where it exists (Mercor US$500M, Handshake AI US$80M, micro1 US$45M, Snorkel US$20.8M, Arena and Proximal none, Appen, TaskUs and Innodata from their H1 or Q2 2025 reports scaled to today's basis, Surge US$1.2bn, Scale's data business US$850M, Turing US$337.5M, Fleet US$1M, Labelbox one fifth of today's, RWS today's divided by 1.45), and every other vendor at today's run-rate divided by 1.5 (the rounded median, 1.54, of the 12 documented multiples); each vendor's split across the five parts is held at today's. The frame is an estimate and its uncertainty is not in the share ranges. Share 1.473% (range 0.446-3.044%). The current share is 7.423% on the same basis, so the change is +6.0 points."},"evaluation":{"low":0.273,"central":0.906,"high":1.883,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"August 2025 run-rate (Q3 2025 frame)","periodEnd":"2025-08-31","source":{"url":"https://sacra.com/research/handshake-vs-mercor/","title":"Handshake vs Mercor","publisher":"Sacra","publishedDate":"2025-09-17","accessedDate":"2026-10-22","tier":"T5","locator":"Sacra estimates Handshake hit $280M in annualized revenue as of August 2025: about $200M from the core job board and $80M from Handshake AI"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Prior share in the Q3 2025 frame. Sacra (T4, 17 Sep 2025) estimates Handshake AI at about US$80M annualised in August 2025, with the tier's 40% range (US$48-112M). Handshake AI independent model evaluation: US$2.4m (0.72-5.04m) = prior run-rate US$80m (48-112m) × 3% (split and the ±50% variation of the smaller parts held as today). Prior part total US$265m = the sum over all vendors of Q3 2025 run-rate × split, plus the allowance of 20 ÷ 1.5 = 13.3. Q3 2025 frame: each vendor's run-rate at the time, documented where it exists (Mercor US$500M, Handshake AI US$80M, micro1 US$45M, Snorkel US$20.8M, Arena and Proximal none, Appen, TaskUs and Innodata from their H1 or Q2 2025 reports scaled to today's basis, Surge US$1.2bn, Scale's data business US$850M, Turing US$337.5M, Fleet US$1M, Labelbox one fifth of today's, RWS today's divided by 1.45), and every other vendor at today's run-rate divided by 1.5 (the rounded median, 1.54, of the 12 documented multiples); each vendor's split across the five parts is held at today's. The frame is an estimate and its uncertainty is not in the share ranges. Share 0.906% (range 0.273-1.883%). The current share is 5.377% on the same basis, so the change is +4.5 points."}},"activityGrowth":{"current":{"low":1000000000,"central":1000000000,"high":1000000000,"unit":"USD","basis":"Gross annualised revenue run-rate in US dollars, before payments to contractors (latest month × 12 or latest quarter × 4)","period":"August 2026 gross run-rate (Handshake AI unit)","periodEnd":"2026-08-31","source":{"url":"https://time.com/collection/time100-ai/2026/garrett-lord/","title":"Garrett Lord: The 100 Most Influential People in AI 2026","publisher":"TIME","publishedDate":"2026-08-27","accessedDate":"2026-10-22","tier":"T2","locator":"The company says (CEO Garrett Lord, quoted by TIME): since launching the AI unit it has grown from zero to more than $1 billion in gross annualized revenue in 16 months, working with eight frontier AI labs"},"note":"The company says (CEO Lord, TIME, 27 Aug 2026) the AI unit has more than US$1bn in gross annualised revenue; the exact figure is not given."},"prior":{"low":48000000,"central":80000000,"high":112000000,"unit":"USD","basis":"Gross annualised revenue run-rate in US dollars, before payments to contractors (latest month × 12 or latest quarter × 4)","period":"August 2025 gross run-rate (Handshake AI unit)","periodEnd":"2025-08-31","source":{"url":"https://sacra.com/research/handshake-vs-mercor/","title":"Handshake vs Mercor","publisher":"Sacra","publishedDate":"2025-09-17","accessedDate":"2026-10-22","tier":"T4","locator":"Sacra estimates Handshake hit $280M in annualized revenue as of August 2025: about $200M from the core job board and $80M from Handshake AI"},"note":"Sacra (17 Sep 2025, T4) estimates Handshake AI at about US$80M annualised as of August 2025 (the total of US$280M less about US$200M from the job board); T4 half-width 40%. Growth is about 12.5 times, above the engine's +300% cap."}},"baselineRevenueUsd":{"low":1000000000,"central":1000000000,"high":1000000000,"unit":"USD","basis":"Gross annualised revenue run-rate in US dollars, before payments to contractors (latest month × 12 or latest quarter × 4)","period":"August 2026 gross run-rate (Handshake AI unit)","periodEnd":"2026-08-31","source":{"url":"https://time.com/collection/time100-ai/2026/garrett-lord/","title":"Garrett Lord: The 100 Most Influential People in AI 2026","publisher":"TIME","publishedDate":"2026-08-27","accessedDate":"2026-10-22","tier":"T2","locator":"The company says (CEO Garrett Lord, quoted by TIME): since launching the AI unit it has grown from zero to more than $1 billion in gross annualized revenue in 16 months, working with eight frontier AI labs"},"note":"The company says (CEO Lord, TIME) the AI unit has more than US$1bn in gross annualised revenue; annual revenue is not published. Unit only, without the recruiting business."},"screening":{"expert-data":{"levelLow":1,"levelHigh":3},"rl-environments":{"levelLow":1,"levelHigh":2},"evaluation":{"levelLow":1,"levelHigh":2}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"2d8bd7620b7da275a2390571b3dc5a10f881f839936f13c90b35d819113c062d"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"2d8bd7620b7da275a2390571b3dc5a10f881f839936f13c90b35d819113c062d"}}},"micro1":{"schema":"5lc-index-inputs/1","key":"models-data-evaluation","entityId":"micro1","name":"micro1","kind":"private","commercialActivityVerified":true,"shares":{"expert-data":{"low":5.852,"central":11.998,"high":15.114,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"Annualised gross run-rate, end of Q3 2026 (company statement of 12 Oct 2026)","periodEnd":"2026-09-30","source":{"url":"https://www.micro1.ai/blog/q3","title":"micro1 end of Q3 update","publisher":"micro1","publishedDate":"2026-10-12","accessedDate":"2026-10-22","tier":"T5","locator":"The company says: \"we've surpassed $1B annual run rate while remaining historically profitable\", after \"surpassing $500M run rate this summer\""},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Method: The company says (end-of-Q3 update, 12 Oct 2026) it has 'surpassed $1B annual run rate', after 'surpassing $500M run rate this summer'. The summer figure matches TechCrunch (20 Aug 2026, T3, one anonymous source), which calls it a gross run-rate up from US$100M eight months earlier, so the October figure is read as gross too (the post does not say). Sacra (T4) also puts US$1bn in October 2026. Stated choice: central US$1.0bn (the company's latest statement); low US$0.5bn keeps the independently reported August level in case the October figure rests on a different basis; high US$1.2bn. Gross before contributor pay (net would be US$150-200M at the August level and is not used). Split follows the card notes. Main input: micro1 (T2). Share = company ÷ function total; low and high move only this company's own range and hold the others at central. micro1 expert data for post-training: US$850m (388-1,110m) = run-rate US$1,000m (500-1,200m) × 85% (the smaller parts of a mixed vendor are varied by ±50% and the largest part absorbs the difference). Function total, annualised US$m (central): 7,084, made of identified candidates 6,934 and an allowance of 150 for vendors outside the register (for example Deccan AI, Shaip, Cogito Tech, Sigma AI, Centaur Labs, Revelo, e2f, Patronus AI, Haize Labs, Lightly, Mostly AI, Abundant, Ethos, Datumo and smaller environment builders). Largest, US$m: mercor 1,650, surge-ai 1,296, handshake-ai 882, micro1 850, scale-ai 412, other candidates 1,844, vendors outside the register 150. Full table: a local evidence capture. Share 11.998% (range 5.852-15.114%). Source published date: 2026-10-12 (full date)."},"labelling-multimodal":{"participates":false,"evidence":{"url":"https://www.micro1.ai/","title":"micro1 homepage","publisher":"micro1","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T2","locator":"Offerings: expert human data, Realm RL environments and evaluations, Cortex evaluation platform, robotics data; no large-scale labelling and multimodal data offering described"}},"rl-environments":{"low":1.866,"central":7.07,"high":12.044,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"Annualised gross run-rate, end of Q3 2026 (company statement of 12 Oct 2026)","periodEnd":"2026-09-30","source":{"url":"https://www.micro1.ai/blog/q3","title":"micro1 end of Q3 update","publisher":"micro1","publishedDate":"2026-10-12","accessedDate":"2026-10-22","tier":"T5","locator":"The company says: \"we've surpassed $1B annual run rate while remaining historically profitable\", after \"surpassing $500M run rate this summer\""},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Method: The company says (end-of-Q3 update, 12 Oct 2026) it has 'surpassed $1B annual run rate', after 'surpassing $500M run rate this summer'. The summer figure matches TechCrunch (20 Aug 2026, T3, one anonymous source), which calls it a gross run-rate up from US$100M eight months earlier, so the October figure is read as gross too (the post does not say). Sacra (T4) also puts US$1bn in October 2026. Stated choice: central US$1.0bn (the company's latest statement); low US$0.5bn keeps the independently reported August level in case the October figure rests on a different basis; high US$1.2bn. Gross before contributor pay (net would be US$150-200M at the August level and is not used). Split follows the card notes. Main input: micro1 (T2). Share = company ÷ function total; low and high move only this company's own range and hold the others at central. micro1 RL environments and agent tasks: US$130m (32.5-234m) = run-rate US$1,000m (500-1,200m) × 13% (the smaller parts of a mixed vendor are varied by ±50% and the largest part absorbs the difference). Function total, annualised US$m (central): 1,839, made of identified candidates 1,739 and an allowance of 100 for vendors outside the register (for example Deccan AI, Shaip, Cogito Tech, Sigma AI, Centaur Labs, Revelo, e2f, Patronus AI, Haize Labs, Lightly, Mostly AI, Abundant, Ethos, Datumo and smaller environment builders). Largest, US$m: mercor 484, surge-ai 414, proximal 140, handshake-ai 136, micro1 130, other candidates 434, vendors outside the register 100. Full table: a local evidence capture. Share 7.07% (range 1.866-12.044%). Source published date: 2026-10-12 (full date)."},"evaluation":{"low":0.876,"central":3.414,"high":5.981,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"Annualised gross run-rate, end of Q3 2026 (company statement of 12 Oct 2026)","periodEnd":"2026-09-30","source":{"url":"https://www.micro1.ai/blog/q3","title":"micro1 end of Q3 update","publisher":"micro1","publishedDate":"2026-10-12","accessedDate":"2026-10-22","tier":"T5","locator":"The company says: \"we've surpassed $1B annual run rate while remaining historically profitable\", after \"surpassing $500M run rate this summer\""},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Method: The company says (end-of-Q3 update, 12 Oct 2026) it has 'surpassed $1B annual run rate', after 'surpassing $500M run rate this summer'. The summer figure matches TechCrunch (20 Aug 2026, T3, one anonymous source), which calls it a gross run-rate up from US$100M eight months earlier, so the October figure is read as gross too (the post does not say). Sacra (T4) also puts US$1bn in October 2026. Stated choice: central US$1.0bn (the company's latest statement); low US$0.5bn keeps the independently reported August level in case the October figure rests on a different basis; high US$1.2bn. Gross before contributor pay (net would be US$150-200M at the August level and is not used). Split follows the card notes. Main input: micro1 (T2). Share = company ÷ function total; low and high move only this company's own range and hold the others at central. micro1 independent model evaluation: US$20m (5-36m) = run-rate US$1,000m (500-1,200m) × 2% (the smaller parts of a mixed vendor are varied by ±50% and the largest part absorbs the difference). Function total, annualised US$m (central): 586, made of identified candidates 566 and an allowance of 20 for vendors outside the register (for example Deccan AI, Shaip, Cogito Tech, Sigma AI, Centaur Labs, Revelo, e2f, Patronus AI, Haize Labs, Lightly, Mostly AI, Abundant, Ethos, Datumo and smaller environment builders). Largest, US$m: arena-ai 110, surge-ai 90, mercor 66, scale-ai 45, innodata 33, other candidates 222, vendors outside the register 20. Full table: a local evidence capture. Share 3.414% (range 0.876-5.981%). Source published date: 2026-10-12 (full date)."},"data-tooling":{"participates":false,"evidence":{"url":"https://www.micro1.ai/","title":"micro1 homepage","publisher":"micro1","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T2","locator":"Offerings: expert human data, Realm RL environments and evaluations, Cortex evaluation platform, robotics data; no labelling and data software offering described"}}},"priorShares":{"expert-data":{"low":0.913,"central":1.282,"high":1.546,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"September 2025 ARR (Q3 2025 frame)","periodEnd":"2025-09-30","source":{"url":"https://techcrunch.com/2025/12/04/micro1-a-scale-ai-competitor-touts-crossing-100m-arr/","title":"Micro1, a Scale AI competitor, touts crossing $100M ARR","publisher":"TechCrunch","publishedDate":"2025-12-04","accessedDate":"2026-10-22","tier":"T5","locator":"The US$100 million ARR is \"more than double the revenue Micro1 reported in September\" 2025, so September 2025 was below US$50M"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Prior share in the Q3 2025 frame. TechCrunch (4 Dec 2025) says the company's US$100M ARR was more than double what it reported in September 2025, so September was below US$50M; central US$45M, range US$35-50M. micro1 expert data for post-training: US$38.2m (27.1-46.2m) = prior run-rate US$45m (35-50m) × 85% (split and the ±50% variation of the smaller parts held as today). Prior part total US$2,983m = the sum over all vendors of Q3 2025 run-rate × split, plus the allowance of 150 ÷ 1.5 = 100. Q3 2025 frame: each vendor's run-rate at the time, documented where it exists (Mercor US$500M, Handshake AI US$80M, micro1 US$45M, Snorkel US$20.8M, Arena and Proximal none, Appen, TaskUs and Innodata from their H1 or Q2 2025 reports scaled to today's basis, Surge US$1.2bn, Scale's data business US$850M, Turing US$337.5M, Fleet US$1M, Labelbox one fifth of today's, RWS today's divided by 1.45), and every other vendor at today's run-rate divided by 1.5 (the rounded median, 1.54, of the 12 documented multiples); each vendor's split across the five parts is held at today's. The frame is an estimate and its uncertainty is not in the share ranges. Share 1.282% (range 0.913-1.546%). The current share is 11.998% on the same basis, so the change is +10.7 points."},"rl-environments":{"low":0.324,"central":0.829,"high":1.373,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"September 2025 ARR (Q3 2025 frame)","periodEnd":"2025-09-30","source":{"url":"https://techcrunch.com/2025/12/04/micro1-a-scale-ai-competitor-touts-crossing-100m-arr/","title":"Micro1, a Scale AI competitor, touts crossing $100M ARR","publisher":"TechCrunch","publishedDate":"2025-12-04","accessedDate":"2026-10-22","tier":"T5","locator":"The US$100 million ARR is \"more than double the revenue Micro1 reported in September\" 2025, so September 2025 was below US$50M"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Prior share in the Q3 2025 frame. TechCrunch (4 Dec 2025) says the company's US$100M ARR was more than double what it reported in September 2025, so September was below US$50M; central US$45M, range US$35-50M. micro1 RL environments and agent tasks: US$5.85m (2.27-9.75m) = prior run-rate US$45m (35-50m) × 13% (split and the ±50% variation of the smaller parts held as today). Prior part total US$706m = the sum over all vendors of Q3 2025 run-rate × split, plus the allowance of 100 ÷ 1.5 = 66.7. Q3 2025 frame: each vendor's run-rate at the time, documented where it exists (Mercor US$500M, Handshake AI US$80M, micro1 US$45M, Snorkel US$20.8M, Arena and Proximal none, Appen, TaskUs and Innodata from their H1 or Q2 2025 reports scaled to today's basis, Surge US$1.2bn, Scale's data business US$850M, Turing US$337.5M, Fleet US$1M, Labelbox one fifth of today's, RWS today's divided by 1.45), and every other vendor at today's run-rate divided by 1.5 (the rounded median, 1.54, of the 12 documented multiples); each vendor's split across the five parts is held at today's. The frame is an estimate and its uncertainty is not in the share ranges. Share 0.829% (range 0.324-1.373%). The current share is 7.07% on the same basis, so the change is +6.2 points."},"evaluation":{"low":0.132,"central":0.34,"high":0.565,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"September 2025 ARR (Q3 2025 frame)","periodEnd":"2025-09-30","source":{"url":"https://techcrunch.com/2025/12/04/micro1-a-scale-ai-competitor-touts-crossing-100m-arr/","title":"Micro1, a Scale AI competitor, touts crossing $100M ARR","publisher":"TechCrunch","publishedDate":"2025-12-04","accessedDate":"2026-10-22","tier":"T5","locator":"The US$100 million ARR is \"more than double the revenue Micro1 reported in September\" 2025, so September 2025 was below US$50M"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Prior share in the Q3 2025 frame. TechCrunch (4 Dec 2025) says the company's US$100M ARR was more than double what it reported in September 2025, so September was below US$50M; central US$45M, range US$35-50M. micro1 independent model evaluation: US$0.9m (0.35-1.5m) = prior run-rate US$45m (35-50m) × 2% (split and the ±50% variation of the smaller parts held as today). Prior part total US$265m = the sum over all vendors of Q3 2025 run-rate × split, plus the allowance of 20 ÷ 1.5 = 13.3. Q3 2025 frame: each vendor's run-rate at the time, documented where it exists (Mercor US$500M, Handshake AI US$80M, micro1 US$45M, Snorkel US$20.8M, Arena and Proximal none, Appen, TaskUs and Innodata from their H1 or Q2 2025 reports scaled to today's basis, Surge US$1.2bn, Scale's data business US$850M, Turing US$337.5M, Fleet US$1M, Labelbox one fifth of today's, RWS today's divided by 1.45), and every other vendor at today's run-rate divided by 1.5 (the rounded median, 1.54, of the 12 documented multiples); each vendor's split across the five parts is held at today's. The frame is an estimate and its uncertainty is not in the share ranges. Share 0.34% (range 0.132-0.565%). The current share is 3.414% on the same basis, so the change is +3.1 points."}},"activityGrowth":{"current":{"low":1000000000,"central":1000000000,"high":1000000000,"unit":"USD","basis":"Annualised revenue run-rate in US dollars (ARR in December 2025, gross run-rate in 2026; latest month × 12 or latest quarter × 4)","period":"End of Q3 2026 run-rate","periodEnd":"2026-09-30","source":{"url":"https://www.micro1.ai/blog/q3","title":"micro1 end of Q3 update","publisher":"micro1","publishedDate":"2026-10-12","accessedDate":"2026-10-22","tier":"T2","locator":"The company says: \"we've surpassed $1B annual run rate while remaining historically profitable\", after \"surpassing $500M run rate this summer\""},"note":"The company says (12 Oct 2026) it has surpassed a US$1B annual run-rate."},"prior":{"low":100000000,"central":100000000,"high":100000000,"unit":"USD","basis":"Annualised revenue run-rate in US dollars (ARR in December 2025, gross run-rate in 2026; latest month × 12 or latest quarter × 4)","period":"December 2025 ARR","periodEnd":"2025-12-04","source":{"url":"https://techcrunch.com/2025/12/04/micro1-a-scale-ai-competitor-touts-crossing-100m-arr/","title":"Micro1, a Scale AI competitor, touts crossing $100M ARR","publisher":"TechCrunch","publishedDate":"2025-12-04","accessedDate":"2026-10-22","tier":"T2","locator":"The company says (founder and CEO Ali Ansari) it has surpassed $100 million in ARR, more than double the revenue it reported in September 2025; it started 2025 at roughly $7 million"},"note":"The company says (founder Ali Ansari, TechCrunch, 4 Dec 2025) it has surpassed US$100M in ARR; TechCrunch (20 Aug 2026) treats the same US$100M as the starting point of its gross run-rate. Growth is about 10 times (9.9 months), above the engine's +300% cap."}},"baselineRevenueUsd":{"low":1000000000,"central":1000000000,"high":1000000000,"unit":"USD","basis":"Gross annualised revenue run-rate in US dollars, before payments to contractors (latest month × 12 or latest quarter × 4)","period":"End of Q3 2026 run-rate","periodEnd":"2026-09-30","source":{"url":"https://www.micro1.ai/blog/q3","title":"micro1 end of Q3 update","publisher":"micro1","publishedDate":"2026-10-12","accessedDate":"2026-10-22","tier":"T2","locator":"The company says: \"we've surpassed $1B annual run rate while remaining historically profitable\", after \"surpassing $500M run rate this summer\""},"note":"The company says it has surpassed a US$1B annual run-rate (12 Oct 2026); annual revenue is not published."},"screening":{"expert-data":{"levelLow":1,"levelHigh":3},"rl-environments":{"levelLow":1,"levelHigh":2},"evaluation":{"levelLow":1,"levelHigh":2}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"a44c85c119c1aac2a51afc70870fafdf6b9a5bcda808f73335fadca4e34b4690"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"a44c85c119c1aac2a51afc70870fafdf6b9a5bcda808f73335fadca4e34b4690"}}},"scale-ai":{"schema":"5lc-index-inputs/1","key":"models-data-evaluation","entityId":"scale-ai","name":"Scale AI","kind":"private","commercialActivityVerified":true,"shares":{"expert-data":{"low":2.84,"central":5.823,"high":10.407,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"2026 data-business run-rate (estimate)","periodEnd":"2026-09-30","source":{"url":"https://forbes.pe/negocios/2026-05-15/asi-funciona-scale-ai-tras-el-sonado-acuerdo-de-meta-por-us14-000-millones/","title":"Así funciona Scale AI tras el sonado acuerdo de Meta por US$14.000 millones","publisher":"Forbes Perú (republishing Forbes US, 14 May 2026)","publishedDate":"2026-05-15","accessedDate":"2026-10-22","tier":"T5","locator":"Scale says it expects to exceed $1 billion of revenue in 2026; 2025 revenue was close to $1 billion, up from $870 million in 2024; applications annualized revenue reached $200 million by the end of 2025; data labeling is still the large majority of revenue (CEO Droege)"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Method: Data business only (enterprise and government applications are in Applications). The company says 2025 was its biggest year of revenue, the data business is profitable and the applications business more than doubled in the second half of 2025 and should roughly double again in 2026 (blog, 22 Jan 2026). Forbes, republished by Forbes Perú on 15 May 2026 (T3; the forbes.com original returned HTTP 403 at screening and was not retried), reports that Scale expects to exceed US$1bn of revenue in 2026, that 2025 revenue was close to US$1bn (US$870M in 2024), that applications reached US$200M annualised by the end of 2025 and that data labelling is still the vast majority of revenue. Central US$0.75bn = about US$1.05bn of expected 2026 revenue less about US$0.3bn of applications (US$200M annualised at the end of 2025, roughly doubling over 2026); low US$0.6bn allows for the lost OpenAI work after Meta's stake, high US$1.0bn. Split follows the card notes. Standalone software sales are not listed on Scale's data-engine page. Main input: Forbes Perú (republishing Forbes US, 14 May 2026) (T3). Share = company ÷ function total; low and high move only this company's own range and hold the others at central. Scale AI expert data for post-training: US$413m (195-775m) = run-rate US$750m (600-1,000m) × 55% (the smaller parts of a mixed vendor are varied by ±50% and the largest part absorbs the difference). Function total, annualised US$m (central): 7,084, made of identified candidates 6,934 and an allowance of 150 for vendors outside the register (for example Deccan AI, Shaip, Cogito Tech, Sigma AI, Centaur Labs, Revelo, e2f, Patronus AI, Haize Labs, Lightly, Mostly AI, Abundant, Ethos, Datumo and smaller environment builders). Largest, US$m: mercor 1,650, surge-ai 1,296, handshake-ai 882, micro1 850, scale-ai 412, other candidates 1,844, vendors outside the register 150. Full table: a local evidence capture. Share 5.823% (range 2.84-10.407%). Source published date: 2026-05-15 (full date)."},"labelling-multimodal":{"low":6.555,"central":14.921,"high":25.967,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"2026 data-business run-rate (estimate)","periodEnd":"2026-09-30","source":{"url":"https://forbes.pe/negocios/2026-05-15/asi-funciona-scale-ai-tras-el-sonado-acuerdo-de-meta-por-us14-000-millones/","title":"Así funciona Scale AI tras el sonado acuerdo de Meta por US$14.000 millones","publisher":"Forbes Perú (republishing Forbes US, 14 May 2026)","publishedDate":"2026-05-15","accessedDate":"2026-10-22","tier":"T5","locator":"Scale says it expects to exceed $1 billion of revenue in 2026; 2025 revenue was close to $1 billion, up from $870 million in 2024; applications annualized revenue reached $200 million by the end of 2025; data labeling is still the large majority of revenue (CEO Droege)"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Method: Data business only (enterprise and government applications are in Applications). The company says 2025 was its biggest year of revenue, the data business is profitable and the applications business more than doubled in the second half of 2025 and should roughly double again in 2026 (blog, 22 Jan 2026). Forbes, republished by Forbes Perú on 15 May 2026 (T3; the forbes.com original returned HTTP 403 at screening and was not retried), reports that Scale expects to exceed US$1bn of revenue in 2026, that 2025 revenue was close to US$1bn (US$870M in 2024), that applications reached US$200M annualised by the end of 2025 and that data labelling is still the vast majority of revenue. Central US$0.75bn = about US$1.05bn of expected 2026 revenue less about US$0.3bn of applications (US$200M annualised at the end of 2025, roughly doubling over 2026); low US$0.6bn allows for the lost OpenAI work after Meta's stake, high US$1.0bn. Split follows the card notes. Standalone software sales are not listed on Scale's data-engine page. Main input: Forbes Perú (republishing Forbes US, 14 May 2026) (T3). Share = company ÷ function total; low and high move only this company's own range and hold the others at central. Scale AI large-scale labelling and multimodal data: US$262m (105-525m) = run-rate US$750m (600-1,000m) × 35% (the smaller parts of a mixed vendor are varied by ±50% and the largest part absorbs the difference). Function total, annualised US$m (central): 1,759, made of identified candidates 1,609 and an allowance of 150 for vendors outside the register (for example Deccan AI, Shaip, Cogito Tech, Sigma AI, Centaur Labs, Revelo, e2f, Patronus AI, Haize Labs, Lightly, Mostly AI, Abundant, Ethos, Datumo and smaller environment builders). Largest, US$m: scale-ai 262, appen 176, taskus 146, rws-trainai 140, telus-digital 124, other candidates 762, vendors outside the register 150. Full table: a local evidence capture. Share 14.921% (range 6.555-25.967%). Source published date: 2026-05-15 (full date)."},"rl-environments":{"low":0.659,"central":1.631,"high":3.211,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"2026 data-business run-rate (estimate)","periodEnd":"2026-09-30","source":{"url":"https://forbes.pe/negocios/2026-05-15/asi-funciona-scale-ai-tras-el-sonado-acuerdo-de-meta-por-us14-000-millones/","title":"Así funciona Scale AI tras el sonado acuerdo de Meta por US$14.000 millones","publisher":"Forbes Perú (republishing Forbes US, 14 May 2026)","publishedDate":"2026-05-15","accessedDate":"2026-10-22","tier":"T5","locator":"Scale says it expects to exceed $1 billion of revenue in 2026; 2025 revenue was close to $1 billion, up from $870 million in 2024; applications annualized revenue reached $200 million by the end of 2025; data labeling is still the large majority of revenue (CEO Droege)"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Method: Data business only (enterprise and government applications are in Applications). The company says 2025 was its biggest year of revenue, the data business is profitable and the applications business more than doubled in the second half of 2025 and should roughly double again in 2026 (blog, 22 Jan 2026). Forbes, republished by Forbes Perú on 15 May 2026 (T3; the forbes.com original returned HTTP 403 at screening and was not retried), reports that Scale expects to exceed US$1bn of revenue in 2026, that 2025 revenue was close to US$1bn (US$870M in 2024), that applications reached US$200M annualised by the end of 2025 and that data labelling is still the vast majority of revenue. Central US$0.75bn = about US$1.05bn of expected 2026 revenue less about US$0.3bn of applications (US$200M annualised at the end of 2025, roughly doubling over 2026); low US$0.6bn allows for the lost OpenAI work after Meta's stake, high US$1.0bn. Split follows the card notes. Standalone software sales are not listed on Scale's data-engine page. Main input: Forbes Perú (republishing Forbes US, 14 May 2026) (T3). Share = company ÷ function total; low and high move only this company's own range and hold the others at central. Scale AI RL environments and agent tasks: US$30m (12-60m) = run-rate US$750m (600-1,000m) × 4% (the smaller parts of a mixed vendor are varied by ±50% and the largest part absorbs the difference). Function total, annualised US$m (central): 1,839, made of identified candidates 1,739 and an allowance of 100 for vendors outside the register (for example Deccan AI, Shaip, Cogito Tech, Sigma AI, Centaur Labs, Revelo, e2f, Patronus AI, Haize Labs, Lightly, Mostly AI, Abundant, Ethos, Datumo and smaller environment builders). Largest, US$m: mercor 484, surge-ai 414, proximal 140, handshake-ai 136, micro1 130, other candidates 434, vendors outside the register 100. Full table: a local evidence capture. Share 1.631% (range 0.659-3.211%). Source published date: 2026-05-15 (full date)."},"evaluation":{"low":3.221,"central":7.681,"high":14.266,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"2026 data-business run-rate (estimate)","periodEnd":"2026-09-30","source":{"url":"https://forbes.pe/negocios/2026-05-15/asi-funciona-scale-ai-tras-el-sonado-acuerdo-de-meta-por-us14-000-millones/","title":"Así funciona Scale AI tras el sonado acuerdo de Meta por US$14.000 millones","publisher":"Forbes Perú (republishing Forbes US, 14 May 2026)","publishedDate":"2026-05-15","accessedDate":"2026-10-22","tier":"T5","locator":"Scale says it expects to exceed $1 billion of revenue in 2026; 2025 revenue was close to $1 billion, up from $870 million in 2024; applications annualized revenue reached $200 million by the end of 2025; data labeling is still the large majority of revenue (CEO Droege)"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Method: Data business only (enterprise and government applications are in Applications). The company says 2025 was its biggest year of revenue, the data business is profitable and the applications business more than doubled in the second half of 2025 and should roughly double again in 2026 (blog, 22 Jan 2026). Forbes, republished by Forbes Perú on 15 May 2026 (T3; the forbes.com original returned HTTP 403 at screening and was not retried), reports that Scale expects to exceed US$1bn of revenue in 2026, that 2025 revenue was close to US$1bn (US$870M in 2024), that applications reached US$200M annualised by the end of 2025 and that data labelling is still the vast majority of revenue. Central US$0.75bn = about US$1.05bn of expected 2026 revenue less about US$0.3bn of applications (US$200M annualised at the end of 2025, roughly doubling over 2026); low US$0.6bn allows for the lost OpenAI work after Meta's stake, high US$1.0bn. Split follows the card notes. Standalone software sales are not listed on Scale's data-engine page. Main input: Forbes Perú (republishing Forbes US, 14 May 2026) (T3). Share = company ÷ function total; low and high move only this company's own range and hold the others at central. Scale AI independent model evaluation: US$45m (18-90m) = run-rate US$750m (600-1,000m) × 6% (the smaller parts of a mixed vendor are varied by ±50% and the largest part absorbs the difference). Function total, annualised US$m (central): 586, made of identified candidates 566 and an allowance of 20 for vendors outside the register (for example Deccan AI, Shaip, Cogito Tech, Sigma AI, Centaur Labs, Revelo, e2f, Patronus AI, Haize Labs, Lightly, Mostly AI, Abundant, Ethos, Datumo and smaller environment builders). Largest, US$m: arena-ai 110, surge-ai 90, mercor 66, scale-ai 45, innodata 33, other candidates 222, vendors outside the register 20. Full table: a local evidence capture. Share 7.681% (range 3.221-14.266%). Source published date: 2026-05-15 (full date)."},"data-tooling":{"participates":false,"evidence":{"url":"https://scale.com/data-engine","title":"Scale Generative AI Data Engine","publisher":"Scale AI","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T2","locator":"Offerings: prompt-response creation, RLHF, red teaming, evaluation and multimodal annotation (text, image, video, 3D LiDAR); no standalone software sales listed; no labelling and data software offering described"}}},"activityGrowth":{"current":{"low":900000000,"central":950000000,"high":1000000000,"unit":"USD","basis":"Total company revenue, calendar year, US dollars (includes enterprise and government applications, outside this sublayer)","period":"CY2025","periodEnd":"2025-12-31","source":{"url":"https://forbes.pe/negocios/2026-05-15/asi-funciona-scale-ai-tras-el-sonado-acuerdo-de-meta-por-us14-000-millones/","title":"Así funciona Scale AI tras el sonado acuerdo de Meta por US$14.000 millones","publisher":"Forbes Perú (republishing Forbes US, 14 May 2026)","publishedDate":"2026-05-15","accessedDate":"2026-10-22","tier":"T3","locator":"Scale says it expects to exceed $1 billion of revenue in 2026; 2025 revenue was close to $1 billion, up from $870 million in 2024; applications annualized revenue reached $200 million by the end of 2025; data labeling is still the large majority of revenue (CEO Droege)"},"note":"Forbes (as republished by Forbes Perú) says 2025 revenue was \"close to\" US$1bn, so central US$0.95bn (US$0.9-1.0bn). The total includes applications, which more than doubled in H2 2025, so growth overstates the data business; no prior shares are given because the current data-business figure is an estimate."},"prior":{"low":870000000,"central":870000000,"high":870000000,"unit":"USD","basis":"Total company revenue, calendar year, US dollars (includes enterprise and government applications, outside this sublayer)","period":"CY2024","periodEnd":"2024-12-31","source":{"url":"https://siliconangle.com/2025/07/01/data-labeling-startup-surge-ai-seeking-1b-first-capital-raise-reports-say/","title":"Data labeling startup Surge AI seeking $1B in first capital raise, reports say","publisher":"SiliconANGLE (reporting Reuters)","publishedDate":"2025-07-01","accessedDate":"2026-10-22","tier":"T3","locator":"Scale generated $870 million in sales the year before (Reuters, anonymous sources)"},"note":"Scale generated US$870M in sales in 2024 (Reuters via SiliconANGLE, 1 Jul 2025, T3); Forbes (as republished by Forbes Perú) also says 2025 revenue was up from US$870M. A conflicting T4 figure exists: SemiAnalysis (6 Jan 2026), citing Sacra and itself, says revenue was \"north of $1.4B in 2024\"; the two T3 reports agree and are used, but on the T4 figure 2025 would be a decline."}},"baselineRevenueUsd":{"low":900000000,"central":950000000,"high":1000000000,"unit":"USD","basis":"Total company revenue, calendar year, US dollars (includes enterprise and government applications, outside this sublayer)","period":"CY2025","periodEnd":"2025-12-31","offBasis":true,"source":{"url":"https://forbes.pe/negocios/2026-05-15/asi-funciona-scale-ai-tras-el-sonado-acuerdo-de-meta-por-us14-000-millones/","title":"Así funciona Scale AI tras el sonado acuerdo de Meta por US$14.000 millones","publisher":"Forbes Perú (republishing Forbes US, 14 May 2026)","publishedDate":"2026-05-15","accessedDate":"2026-10-22","tier":"T3","locator":"Scale says it expects to exceed $1 billion of revenue in 2026; 2025 revenue was close to $1 billion, up from $870 million in 2024; applications annualized revenue reached $200 million by the end of 2025; data labeling is still the large majority of revenue (CEO Droege)"},"note":"Forbes (via Forbes Perú): 2025 revenue was close to US$1bn. The data business alone is not published, so this total includes applications and is flagged off basis."},"screening":{"expert-data":{"levelLow":1,"levelHigh":3},"labelling-multimodal":{"levelLow":1,"levelHigh":2},"rl-environments":{"levelLow":1,"levelHigh":2},"evaluation":{"levelLow":1,"levelHigh":2}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"1191f214df4cc5adca7c554f3dd2793c8426cd23f63cd0e42c0f9aab1b5d20e2"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"1191f214df4cc5adca7c554f3dd2793c8426cd23f63cd0e42c0f9aab1b5d20e2"}}},"arena-ai":{"schema":"5lc-index-inputs/1","key":"models-data-evaluation","entityId":"arena-ai","name":"Arena","kind":"private","commercialActivityVerified":true,"shares":{"expert-data":{"participates":false,"evidence":{"url":"https://arena.ai/blog/arena-100m-revenue","title":"Arena Reaches $100M in 8 Months","publisher":"Arena","publishedDate":"2026-06-29","accessedDate":"2026-10-22","tier":"T2","locator":"The company says Arena crossed $100M annualized revenue run rate within eight months of launching its enterprise offering; works with AI labs to evaluate their models; no expert data for post-training offering described"}},"labelling-multimodal":{"participates":false,"evidence":{"url":"https://arena.ai/blog/arena-100m-revenue","title":"Arena Reaches $100M in 8 Months","publisher":"Arena","publishedDate":"2026-06-29","accessedDate":"2026-10-22","tier":"T2","locator":"The company says Arena crossed $100M annualized revenue run rate within eight months of launching its enterprise offering; works with AI labs to evaluate their models; no large-scale labelling and multimodal data offering described"}},"rl-environments":{"participates":false,"evidence":{"url":"https://arena.ai/blog/arena-100m-revenue","title":"Arena Reaches $100M in 8 Months","publisher":"Arena","publishedDate":"2026-06-29","accessedDate":"2026-10-22","tier":"T2","locator":"The company says Arena crossed $100M annualized revenue run rate within eight months of launching its enterprise offering; works with AI labs to evaluate their models; no RL environments and agent tasks offering described"}},"evaluation":{"low":17.365,"central":18.775,"high":23.966,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"Annualised run-rate, June 2026","periodEnd":"2026-06-30","source":{"url":"https://arena.ai/blog/arena-100m-revenue","title":"Arena Reaches $100M in 8 Months","publisher":"Arena","publishedDate":"2026-06-29","accessedDate":"2026-10-22","tier":"T5","locator":"The company says: Arena has crossed $100M annualized revenue run rate within eight months of launching our enterprise offering (CEO letter)"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Method: The company says (29 Jun 2026) Arena crossed a US$100M annualised revenue run-rate within eight months of launching its enterprise offering, selling evaluations to AI labs and enterprises. TechCrunch (29 Jun 2026, T3) says it began generating revenue in September 2025, charges for consumption (so the revenue is not recurring), and that the run-rate was US$30M in December 2025 (6 Jan 2026). Central US$110M (a further quarter of growth), low US$100M, high US$150M. All revenue is treated as independent model evaluation; its blog describes evaluation only, so the other four functions are declared not participating. Main input: Arena (T2). Share = company ÷ function total; low and high move only this company's own range and hold the others at central. Arena independent model evaluation: US$110m (100-150m) = run-rate US$110m (100-150m) × 100%. Function total, annualised US$m (central): 586, made of identified candidates 566 and an allowance of 20 for vendors outside the register (for example Deccan AI, Shaip, Cogito Tech, Sigma AI, Centaur Labs, Revelo, e2f, Patronus AI, Haize Labs, Lightly, Mostly AI, Abundant, Ethos, Datumo and smaller environment builders). Largest, US$m: arena-ai 110, surge-ai 90, mercor 66, scale-ai 45, innodata 33, other candidates 222, vendors outside the register 20. Full table: a local evidence capture. Share 18.775% (range 17.365-23.966%). Source published date: 2026-06-29 (full date)."},"data-tooling":{"participates":false,"evidence":{"url":"https://arena.ai/blog/arena-100m-revenue","title":"Arena Reaches $100M in 8 Months","publisher":"Arena","publishedDate":"2026-06-29","accessedDate":"2026-10-22","tier":"T2","locator":"The company says Arena crossed $100M annualized revenue run rate within eight months of launching its enterprise offering; works with AI labs to evaluate their models; no labelling and data software offering described"}}},"priorShares":{"evaluation":{"low":0,"central":0,"high":0,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"August 2025, before the first commercial launch","periodEnd":"2025-08-31","source":{"url":"https://techcrunch.com/2026/06/29/arena-the-ai-leaderboard-everyone-uses-is-now-a-100m-business/","title":"Arena, the AI leaderboard everyone uses, is now a $100M business","publisher":"TechCrunch","publishedDate":"2026-06-29","accessedDate":"2026-10-22","tier":"T5","locator":"Arena \"began generating revenue from its platform in September\" (2025) when it introduced AI Evaluations"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Prior share in the Q3 2025 frame. Zero: the company began earning revenue in September 2025 (TechCrunch, 29 Jun 2026; first commercial service launched that month, TechCrunch 6 Jan 2026), so it had no share in August 2025. Arena independent model evaluation: US$0m (0-0m) = prior run-rate US$0m (0-0m) × 100%. Prior part total US$265m = the sum over all vendors of Q3 2025 run-rate × split, plus the allowance of 20 ÷ 1.5 = 13.3. Q3 2025 frame: each vendor's run-rate at the time, documented where it exists (Mercor US$500M, Handshake AI US$80M, micro1 US$45M, Snorkel US$20.8M, Arena and Proximal none, Appen, TaskUs and Innodata from their H1 or Q2 2025 reports scaled to today's basis, Surge US$1.2bn, Scale's data business US$850M, Turing US$337.5M, Fleet US$1M, Labelbox one fifth of today's, RWS today's divided by 1.45), and every other vendor at today's run-rate divided by 1.5 (the rounded median, 1.54, of the 12 documented multiples); each vendor's split across the five parts is held at today's. The frame is an estimate and its uncertainty is not in the share ranges. Share 0% (range 0-0%). The current share is 18.775% on the same basis, so the change is +18.8 points."}},"activityGrowth":{"current":{"low":100000000,"central":100000000,"high":100000000,"unit":"USD","basis":"Annualised revenue run-rate in US dollars, consumption-based (latest month × 12 or latest quarter × 4)","period":"June 2026 run-rate","periodEnd":"2026-06-30","source":{"url":"https://arena.ai/blog/arena-100m-revenue","title":"Arena Reaches $100M in 8 Months","publisher":"Arena","publishedDate":"2026-06-29","accessedDate":"2026-10-22","tier":"T2","locator":"The company says: Arena has crossed $100M annualized revenue run rate within eight months of launching our enterprise offering (CEO letter)"},"note":"The company says Arena crossed US$100M annualised (blog and CEO letter, 29 Jun 2026); TechCrunch of the same day reports the same."},"prior":{"low":0,"central":0,"high":0,"unit":"USD","basis":"Annualised revenue run-rate in US dollars, consumption-based (latest month × 12 or latest quarter × 4)","period":"August 2025, before the first commercial launch","periodEnd":"2025-08-31","source":{"url":"https://techcrunch.com/2026/06/29/arena-the-ai-leaderboard-everyone-uses-is-now-a-100m-business/","title":"Arena, the AI leaderboard everyone uses, is now a $100M business","publisher":"TechCrunch","publishedDate":"2026-06-29","accessedDate":"2026-10-22","tier":"T3","locator":"Arena reached $100 million in annualized run-rate revenue; it \"began generating revenue from its platform in September\" when it introduced AI Evaluations; CEO Anastasios Angelopoulos says it charges for consumption, so the revenue is not recurring"},"note":"TechCrunch (29 Jun 2026) says Arena \"began generating revenue from its platform in September\" 2025 when it introduced AI Evaluations; TechCrunch of 6 Jan 2026 says it was originally funded through grants and donations. So there was no commercial revenue in August 2025 (documented, not a missing figure). A prior of zero counts as the +300% cap."}},"baselineRevenueUsd":{"low":100000000,"central":100000000,"high":100000000,"unit":"USD","basis":"Annualised revenue run-rate in US dollars, consumption-based","period":"June 2026 run-rate","periodEnd":"2026-06-30","source":{"url":"https://arena.ai/blog/arena-100m-revenue","title":"Arena Reaches $100M in 8 Months","publisher":"Arena","publishedDate":"2026-06-29","accessedDate":"2026-10-22","tier":"T2","locator":"The company says: Arena has crossed $100M annualized revenue run rate within eight months of launching our enterprise offering (CEO letter)"},"note":"The company says it crossed US$100M annualised; annual revenue is not published."},"screening":{"evaluation":{"levelLow":2,"levelHigh":4}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"c609139c2b3ea8c55fb1981ceab6087d1768c9ada9b9ac96ec80d3e24c935cd7"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"c609139c2b3ea8c55fb1981ceab6087d1768c9ada9b9ac96ec80d3e24c935cd7"}}},"snorkel-ai":{"schema":"5lc-index-inputs/1","key":"models-data-evaluation","entityId":"snorkel-ai","name":"Snorkel AI","kind":"private","commercialActivityVerified":true,"shares":{"expert-data":{"low":3.33,"central":3.97,"high":4.893,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"Annualised run-rate, September 2026","periodEnd":"2026-09-22","source":{"url":"https://techcrunch.com/2026/09/22/snorkel-ai-triples-valuation-to-3-5b-as-demand-for-ai-training-data-booms/","title":"Snorkel AI triples valuation to $3.5B as demand for AI training data booms","publisher":"TechCrunch","publishedDate":"2026-09-22","accessedDate":"2026-10-22","tier":"T5","locator":"The company says its current annualized revenue run rate stands at $375 million, an eighteenfold increase over the last 12 months; payments to its human experts sit in cost of goods sold"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Method: The company says (TechCrunch, 22 Sep 2026) its annualised run-rate is US$375M, an eighteenfold increase over the last 12 months, mostly data-as-a-service; payments to its human experts are carried in cost of goods sold, so the run-rate is before them. Split follows the card notes (no split is published): expert datasets 75%, environments 10%, evaluation 5%, software platform 10%. TechCrunch describes Snorkel as selling RL environments and complete datasets rather than human labour, so it declares no large-scale labelling. Main input: TechCrunch (T2). Share = company ÷ function total; low and high move only this company's own range and hold the others at central. Snorkel AI expert data for post-training: US$281m (234-350m) = run-rate US$375m (375-400m) × 75% (the smaller parts of a mixed vendor are varied by ±50% and the largest part absorbs the difference). Function total, annualised US$m (central): 7,084, made of identified candidates 6,934 and an allowance of 150 for vendors outside the register (for example Deccan AI, Shaip, Cogito Tech, Sigma AI, Centaur Labs, Revelo, e2f, Patronus AI, Haize Labs, Lightly, Mostly AI, Abundant, Ethos, Datumo and smaller environment builders). Largest, US$m: mercor 1,650, surge-ai 1,296, handshake-ai 882, micro1 850, scale-ai 412, other candidates 1,844, vendors outside the register 150. Full table: a local evidence capture. Share 3.97% (range 3.33-4.893%). Source published date: 2026-09-22 (full date)."},"labelling-multimodal":{"participates":false,"evidence":{"url":"https://techcrunch.com/2026/09/22/snorkel-ai-triples-valuation-to-3-5b-as-demand-for-ai-training-data-booms/","title":"Snorkel AI triples valuation to $3.5B as demand for AI training data booms","publisher":"TechCrunch","publishedDate":"2026-09-22","accessedDate":"2026-10-22","tier":"T2","locator":"Snorkel says its current annualized revenue run rate stands at $375 million; the article says it sells RL environments and complete datasets, not human labour; no large-scale labelling and multimodal data offering described"}},"rl-environments":{"low":1.03,"central":2.039,"high":3.223,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"Annualised run-rate, September 2026","periodEnd":"2026-09-22","source":{"url":"https://techcrunch.com/2026/09/22/snorkel-ai-triples-valuation-to-3-5b-as-demand-for-ai-training-data-booms/","title":"Snorkel AI triples valuation to $3.5B as demand for AI training data booms","publisher":"TechCrunch","publishedDate":"2026-09-22","accessedDate":"2026-10-22","tier":"T5","locator":"The company says its current annualized revenue run rate stands at $375 million, an eighteenfold increase over the last 12 months; payments to its human experts sit in cost of goods sold"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Method: The company says (TechCrunch, 22 Sep 2026) its annualised run-rate is US$375M, an eighteenfold increase over the last 12 months, mostly data-as-a-service; payments to its human experts are carried in cost of goods sold, so the run-rate is before them. Split follows the card notes (no split is published): expert datasets 75%, environments 10%, evaluation 5%, software platform 10%. TechCrunch describes Snorkel as selling RL environments and complete datasets rather than human labour, so it declares no large-scale labelling. Main input: TechCrunch (T2). Share = company ÷ function total; low and high move only this company's own range and hold the others at central. Snorkel AI RL environments and agent tasks: US$37.5m (18.8-60m) = run-rate US$375m (375-400m) × 10% (the smaller parts of a mixed vendor are varied by ±50% and the largest part absorbs the difference). Function total, annualised US$m (central): 1,839, made of identified candidates 1,739 and an allowance of 100 for vendors outside the register (for example Deccan AI, Shaip, Cogito Tech, Sigma AI, Centaur Labs, Revelo, e2f, Patronus AI, Haize Labs, Lightly, Mostly AI, Abundant, Ethos, Datumo and smaller environment builders). Largest, US$m: mercor 484, surge-ai 414, proximal 140, handshake-ai 136, micro1 130, other candidates 434, vendors outside the register 100. Full table: a local evidence capture. Share 2.039% (range 1.03-3.223%). Source published date: 2026-09-22 (full date)."},"evaluation":{"low":1.626,"central":3.2,"high":5.024,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"Annualised run-rate, September 2026","periodEnd":"2026-09-22","source":{"url":"https://techcrunch.com/2026/09/22/snorkel-ai-triples-valuation-to-3-5b-as-demand-for-ai-training-data-booms/","title":"Snorkel AI triples valuation to $3.5B as demand for AI training data booms","publisher":"TechCrunch","publishedDate":"2026-09-22","accessedDate":"2026-10-22","tier":"T5","locator":"The company says its current annualized revenue run rate stands at $375 million, an eighteenfold increase over the last 12 months; payments to its human experts sit in cost of goods sold"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Method: The company says (TechCrunch, 22 Sep 2026) its annualised run-rate is US$375M, an eighteenfold increase over the last 12 months, mostly data-as-a-service; payments to its human experts are carried in cost of goods sold, so the run-rate is before them. Split follows the card notes (no split is published): expert datasets 75%, environments 10%, evaluation 5%, software platform 10%. TechCrunch describes Snorkel as selling RL environments and complete datasets rather than human labour, so it declares no large-scale labelling. Main input: TechCrunch (T2). Share = company ÷ function total; low and high move only this company's own range and hold the others at central. Snorkel AI independent model evaluation: US$18.8m (9.38-30m) = run-rate US$375m (375-400m) × 5% (the smaller parts of a mixed vendor are varied by ±50% and the largest part absorbs the difference). Function total, annualised US$m (central): 586, made of identified candidates 566 and an allowance of 20 for vendors outside the register (for example Deccan AI, Shaip, Cogito Tech, Sigma AI, Centaur Labs, Revelo, e2f, Patronus AI, Haize Labs, Lightly, Mostly AI, Abundant, Ethos, Datumo and smaller environment builders). Largest, US$m: arena-ai 110, surge-ai 90, mercor 66, scale-ai 45, innodata 33, other candidates 222, vendors outside the register 20. Full table: a local evidence capture. Share 3.2% (range 1.626-5.024%). Source published date: 2026-09-22 (full date)."},"data-tooling":{"low":7.656,"central":14.223,"high":20.968,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"Annualised run-rate, September 2026","periodEnd":"2026-09-22","source":{"url":"https://techcrunch.com/2026/09/22/snorkel-ai-triples-valuation-to-3-5b-as-demand-for-ai-training-data-booms/","title":"Snorkel AI triples valuation to $3.5B as demand for AI training data booms","publisher":"TechCrunch","publishedDate":"2026-09-22","accessedDate":"2026-10-22","tier":"T5","locator":"The company says its current annualized revenue run rate stands at $375 million, an eighteenfold increase over the last 12 months; payments to its human experts sit in cost of goods sold"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Method: The company says (TechCrunch, 22 Sep 2026) its annualised run-rate is US$375M, an eighteenfold increase over the last 12 months, mostly data-as-a-service; payments to its human experts are carried in cost of goods sold, so the run-rate is before them. Split follows the card notes (no split is published): expert datasets 75%, environments 10%, evaluation 5%, software platform 10%. TechCrunch describes Snorkel as selling RL environments and complete datasets rather than human labour, so it declares no large-scale labelling. Main input: TechCrunch (T2). Share = company ÷ function total; low and high move only this company's own range and hold the others at central. Snorkel AI labelling and data software: US$37.5m (18.8-60m) = run-rate US$375m (375-400m) × 10% (the smaller parts of a mixed vendor are varied by ±50% and the largest part absorbs the difference). Function total, annualised US$m (central): 264, made of identified candidates 234 and an allowance of 30 for vendors outside the register (for example Deccan AI, Shaip, Cogito Tech, Sigma AI, Centaur Labs, Revelo, e2f, Patronus AI, Haize Labs, Lightly, Mostly AI, Abundant, Ethos, Datumo and smaller environment builders). Largest, US$m: snorkel-ai 37.5, labelbox 37.5, encord 36, roboflow 25, superannotate 21, other candidates 76.7, vendors outside the register 30. Full table: a local evidence capture. Share 14.223% (range 7.656-20.968%). Source published date: 2026-09-22 (full date)."}},"priorShares":{"expert-data":{"low":0.425,"central":0.524,"high":0.628,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"September 2025 run-rate (Q3 2025 frame)","periodEnd":"2025-09-22","source":{"url":"https://techcrunch.com/2026/09/22/snorkel-ai-triples-valuation-to-3-5b-as-demand-for-ai-training-data-booms/","title":"Snorkel AI triples valuation to $3.5B as demand for AI training data booms","publisher":"TechCrunch","publishedDate":"2026-09-22","accessedDate":"2026-10-22","tier":"T5","locator":"The company says its current annualized revenue run rate stands at $375 million, an eighteenfold increase over the last 12 months; payments to its human experts sit in cost of goods sold"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Prior share in the Q3 2025 frame. The company says its run-rate is up eighteenfold in twelve months (TechCrunch, 22 Sep 2026), so in September 2025 it was about US$375M ÷ 18 = US$20.8M (US$20.3-21.4M). Snorkel AI expert data for post-training: US$15.6m (12.7-18.8m) = prior run-rate US$20.8m (20.3-21.4m) × 75% (split and the ±50% variation of the smaller parts held as today). Prior part total US$2,983m = the sum over all vendors of Q3 2025 run-rate × split, plus the allowance of 150 ÷ 1.5 = 100. Q3 2025 frame: each vendor's run-rate at the time, documented where it exists (Mercor US$500M, Handshake AI US$80M, micro1 US$45M, Snorkel US$20.8M, Arena and Proximal none, Appen, TaskUs and Innodata from their H1 or Q2 2025 reports scaled to today's basis, Surge US$1.2bn, Scale's data business US$850M, Turing US$337.5M, Fleet US$1M, Labelbox one fifth of today's, RWS today's divided by 1.45), and every other vendor at today's run-rate divided by 1.5 (the rounded median, 1.54, of the 12 documented multiples); each vendor's split across the five parts is held at today's. The frame is an estimate and its uncertainty is not in the share ranges. Share 0.524% (range 0.425-0.628%). The current share is 3.97% on the same basis, so the change is +3.4 points."},"rl-environments":{"low":0.144,"central":0.295,"high":0.455,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"September 2025 run-rate (Q3 2025 frame)","periodEnd":"2025-09-22","source":{"url":"https://techcrunch.com/2026/09/22/snorkel-ai-triples-valuation-to-3-5b-as-demand-for-ai-training-data-booms/","title":"Snorkel AI triples valuation to $3.5B as demand for AI training data booms","publisher":"TechCrunch","publishedDate":"2026-09-22","accessedDate":"2026-10-22","tier":"T5","locator":"The company says its current annualized revenue run rate stands at $375 million, an eighteenfold increase over the last 12 months; payments to its human experts sit in cost of goods sold"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Prior share in the Q3 2025 frame. The company says its run-rate is up eighteenfold in twelve months (TechCrunch, 22 Sep 2026), so in September 2025 it was about US$375M ÷ 18 = US$20.8M (US$20.3-21.4M). Snorkel AI RL environments and agent tasks: US$2.08m (1.01-3.21m) = prior run-rate US$20.8m (20.3-21.4m) × 10% (split and the ±50% variation of the smaller parts held as today). Prior part total US$706m = the sum over all vendors of Q3 2025 run-rate × split, plus the allowance of 100 ÷ 1.5 = 66.7. Q3 2025 frame: each vendor's run-rate at the time, documented where it exists (Mercor US$500M, Handshake AI US$80M, micro1 US$45M, Snorkel US$20.8M, Arena and Proximal none, Appen, TaskUs and Innodata from their H1 or Q2 2025 reports scaled to today's basis, Surge US$1.2bn, Scale's data business US$850M, Turing US$337.5M, Fleet US$1M, Labelbox one fifth of today's, RWS today's divided by 1.45), and every other vendor at today's run-rate divided by 1.5 (the rounded median, 1.54, of the 12 documented multiples); each vendor's split across the five parts is held at today's. The frame is an estimate and its uncertainty is not in the share ranges. Share 0.295% (range 0.144-0.455%). The current share is 2.039% on the same basis, so the change is +1.7 points."},"evaluation":{"low":0.192,"central":0.393,"high":0.605,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"September 2025 run-rate (Q3 2025 frame)","periodEnd":"2025-09-22","source":{"url":"https://techcrunch.com/2026/09/22/snorkel-ai-triples-valuation-to-3-5b-as-demand-for-ai-training-data-booms/","title":"Snorkel AI triples valuation to $3.5B as demand for AI training data booms","publisher":"TechCrunch","publishedDate":"2026-09-22","accessedDate":"2026-10-22","tier":"T5","locator":"The company says its current annualized revenue run rate stands at $375 million, an eighteenfold increase over the last 12 months; payments to its human experts sit in cost of goods sold"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Prior share in the Q3 2025 frame. The company says its run-rate is up eighteenfold in twelve months (TechCrunch, 22 Sep 2026), so in September 2025 it was about US$375M ÷ 18 = US$20.8M (US$20.3-21.4M). Snorkel AI independent model evaluation: US$1.04m (0.51-1.61m) = prior run-rate US$20.8m (20.3-21.4m) × 5% (split and the ±50% variation of the smaller parts held as today). Prior part total US$265m = the sum over all vendors of Q3 2025 run-rate × split, plus the allowance of 20 ÷ 1.5 = 13.3. Q3 2025 frame: each vendor's run-rate at the time, documented where it exists (Mercor US$500M, Handshake AI US$80M, micro1 US$45M, Snorkel US$20.8M, Arena and Proximal none, Appen, TaskUs and Innodata from their H1 or Q2 2025 reports scaled to today's basis, Surge US$1.2bn, Scale's data business US$850M, Turing US$337.5M, Fleet US$1M, Labelbox one fifth of today's, RWS today's divided by 1.45), and every other vendor at today's run-rate divided by 1.5 (the rounded median, 1.54, of the 12 documented multiples); each vendor's split across the five parts is held at today's. The frame is an estimate and its uncertainty is not in the share ranges. Share 0.393% (range 0.192-0.605%). The current share is 3.2% on the same basis, so the change is +2.8 points."},"data-tooling":{"low":0.755,"central":1.539,"high":2.355,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"September 2025 run-rate (Q3 2025 frame)","periodEnd":"2025-09-22","source":{"url":"https://techcrunch.com/2026/09/22/snorkel-ai-triples-valuation-to-3-5b-as-demand-for-ai-training-data-booms/","title":"Snorkel AI triples valuation to $3.5B as demand for AI training data booms","publisher":"TechCrunch","publishedDate":"2026-09-22","accessedDate":"2026-10-22","tier":"T5","locator":"The company says its current annualized revenue run rate stands at $375 million, an eighteenfold increase over the last 12 months; payments to its human experts sit in cost of goods sold"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Prior share in the Q3 2025 frame. The company says its run-rate is up eighteenfold in twelve months (TechCrunch, 22 Sep 2026), so in September 2025 it was about US$375M ÷ 18 = US$20.8M (US$20.3-21.4M). Snorkel AI labelling and data software: US$2.08m (1.01-3.21m) = prior run-rate US$20.8m (20.3-21.4m) × 10% (split and the ±50% variation of the smaller parts held as today). Prior part total US$135m = the sum over all vendors of Q3 2025 run-rate × split, plus the allowance of 30 ÷ 1.5 = 20. Q3 2025 frame: each vendor's run-rate at the time, documented where it exists (Mercor US$500M, Handshake AI US$80M, micro1 US$45M, Snorkel US$20.8M, Arena and Proximal none, Appen, TaskUs and Innodata from their H1 or Q2 2025 reports scaled to today's basis, Surge US$1.2bn, Scale's data business US$850M, Turing US$337.5M, Fleet US$1M, Labelbox one fifth of today's, RWS today's divided by 1.45), and every other vendor at today's run-rate divided by 1.5 (the rounded median, 1.54, of the 12 documented multiples); each vendor's split across the five parts is held at today's. The frame is an estimate and its uncertainty is not in the share ranges. Share 1.539% (range 0.755-2.355%). The current share is 14.223% on the same basis, so the change is +12.7 points."}},"activityGrowth":{"current":{"low":375000000,"central":375000000,"high":400000000,"unit":"USD","basis":"Annualised revenue run-rate in US dollars (latest month × 12 or latest quarter × 4)","period":"September 2026 run-rate","periodEnd":"2026-09-22","source":{"url":"https://techcrunch.com/2026/09/22/snorkel-ai-triples-valuation-to-3-5b-as-demand-for-ai-training-data-booms/","title":"Snorkel AI triples valuation to $3.5B as demand for AI training data booms","publisher":"TechCrunch","publishedDate":"2026-09-22","accessedDate":"2026-10-22","tier":"T2","locator":"The company says its current annualized revenue run rate stands at $375 million, an eighteenfold increase over the last 12 months; payments to its human experts sit in cost of goods sold"},"note":"The company says its annualised run-rate stands at US$375M (TechCrunch, 22 Sep 2026); the upper end allows for rounding of \"$375 million\"."},"prior":{"low":20270270.27027027,"central":20833333.333333332,"high":21428571.42857143,"unit":"USD","basis":"Annualised revenue run-rate in US dollars (latest month × 12 or latest quarter × 4)","period":"September 2025 run-rate (derived)","periodEnd":"2025-09-22","source":{"url":"https://techcrunch.com/2026/09/22/snorkel-ai-triples-valuation-to-3-5b-as-demand-for-ai-training-data-booms/","title":"Snorkel AI triples valuation to $3.5B as demand for AI training data booms","publisher":"TechCrunch","publishedDate":"2026-09-22","accessedDate":"2026-10-22","tier":"T5","locator":"The company says its current annualized revenue run rate stands at $375 million, an eighteenfold increase over the last 12 months; payments to its human experts sit in cost of goods sold"},"note":"Derived: the company says the run-rate is an eighteenfold increase over the last 12 months, so a year earlier it was US$375M ÷ 18 = US$20.8M (US$20.3-21.4M if \"eighteenfold\" is rounded from 17.5-18.5). Growth is 18 times, above the engine's +300% cap."}},"baselineRevenueUsd":{"low":375000000,"central":375000000,"high":400000000,"unit":"USD","basis":"Annualised revenue run-rate in US dollars (latest month × 12 or latest quarter × 4)","period":"September 2026 run-rate","periodEnd":"2026-09-22","source":{"url":"https://techcrunch.com/2026/09/22/snorkel-ai-triples-valuation-to-3-5b-as-demand-for-ai-training-data-booms/","title":"Snorkel AI triples valuation to $3.5B as demand for AI training data booms","publisher":"TechCrunch","publishedDate":"2026-09-22","accessedDate":"2026-10-22","tier":"T2","locator":"The company says its current annualized revenue run rate stands at $375 million, an eighteenfold increase over the last 12 months; payments to its human experts sit in cost of goods sold"},"note":"The company says its annualised run-rate is US$375M; annual revenue is not published. Includes the software platform (about 10%)."},"screening":{"expert-data":{"levelLow":1,"levelHigh":3},"rl-environments":{"levelLow":1,"levelHigh":2},"evaluation":{"levelLow":1,"levelHigh":2},"data-tooling":{"levelLow":1,"levelHigh":3}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"54d6a7853fd0356977bdbe33a977a7f5a421ec0f54e9c2783d261939b229047b"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"54d6a7853fd0356977bdbe33a977a7f5a421ec0f54e9c2783d261939b229047b"}}},"turing":{"schema":"5lc-index-inputs/1","key":"models-data-evaluation","entityId":"turing","name":"Turing","kind":"private","commercialActivityVerified":true,"shares":{"expert-data":{"low":1.583,"central":3.458,"high":6.939,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"Annualised revenue, end of 2025 (T4), carried to a 2026 estimate","periodEnd":"2025-12-31","source":{"url":"https://sacra.com/c/turing/","title":"Turing revenue, funding and news","publisher":"Sacra","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T5","locator":"Sacra estimates Turing reached $350M in annualized revenue at the end of 2025, about 17% above the end of 2024 (about $300M); page gives no 2026 figure"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Method: Sacra (T4) estimates US$350M annualised at the end of 2025 (+17% on about US$300M); the company said in January 2025 that it had reached a US$300M ARR (press release, T2). Sacra's page gives no 2026 figure. Turing's own homepage lists datasets, RL environments and an expert network for labs, and Sacra notes 1,000+ enterprise clients of Turing Intelligence. Central US$350M for the in-scope part, low US$200M (part of the revenue is enterprise AI work), high US$600M with 2026 growth. Split follows the card notes. Main input: Sacra (T4). Share = company ÷ function total; low and high move only this company's own range and hold the others at central. Turing expert data for post-training: US$245m (110-510m) = run-rate US$350m (200-600m) × 70% (the smaller parts of a mixed vendor are varied by ±50% and the largest part absorbs the difference). Function total, annualised US$m (central): 7,084, made of identified candidates 6,934 and an allowance of 150 for vendors outside the register (for example Deccan AI, Shaip, Cogito Tech, Sigma AI, Centaur Labs, Revelo, e2f, Patronus AI, Haize Labs, Lightly, Mostly AI, Abundant, Ethos, Datumo and smaller environment builders). Largest, US$m: mercor 1,650, surge-ai 1,296, handshake-ai 882, micro1 850, scale-ai 412, other candidates 1,844, vendors outside the register 150. Full table: a local evidence capture. Share 3.458% (range 1.583-6.939%). Source published date: undated (undated)."},"labelling-multimodal":{"participates":false,"evidence":{"url":"https://www.turing.com/","title":"Turing homepage","publisher":"Turing","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T2","locator":"Offerings: off-the-shelf datasets and RL environments for frontier labs, expert network, enterprise AI; no large-scale labelling and multimodal data offering described"}},"rl-environments":{"low":1.583,"central":5.329,"high":12.645,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"Annualised revenue, end of 2025 (T4), carried to a 2026 estimate","periodEnd":"2025-12-31","source":{"url":"https://sacra.com/c/turing/","title":"Turing revenue, funding and news","publisher":"Sacra","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T5","locator":"Sacra estimates Turing reached $350M in annualized revenue at the end of 2025, about 17% above the end of 2024 (about $300M); page gives no 2026 figure"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Method: Sacra (T4) estimates US$350M annualised at the end of 2025 (+17% on about US$300M); the company said in January 2025 that it had reached a US$300M ARR (press release, T2). Sacra's page gives no 2026 figure. Turing's own homepage lists datasets, RL environments and an expert network for labs, and Sacra notes 1,000+ enterprise clients of Turing Intelligence. Central US$350M for the in-scope part, low US$200M (part of the revenue is enterprise AI work), high US$600M with 2026 growth. Split follows the card notes. Main input: Sacra (T4). Share = company ÷ function total; low and high move only this company's own range and hold the others at central. Turing RL environments and agent tasks: US$98m (28-252m) = run-rate US$350m (200-600m) × 28% (the smaller parts of a mixed vendor are varied by ±50% and the largest part absorbs the difference). Function total, annualised US$m (central): 1,839, made of identified candidates 1,739 and an allowance of 100 for vendors outside the register (for example Deccan AI, Shaip, Cogito Tech, Sigma AI, Centaur Labs, Revelo, e2f, Patronus AI, Haize Labs, Lightly, Mostly AI, Abundant, Ethos, Datumo and smaller environment builders). Largest, US$m: mercor 484, surge-ai 414, proximal 140, handshake-ai 136, micro1 130, other candidates 434, vendors outside the register 100. Full table: a local evidence capture. Share 5.329% (range 1.583-12.645%). Source published date: undated (undated)."},"evaluation":{"low":0.344,"central":1.195,"high":3.016,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"Annualised revenue, end of 2025 (T4), carried to a 2026 estimate","periodEnd":"2025-12-31","source":{"url":"https://sacra.com/c/turing/","title":"Turing revenue, funding and news","publisher":"Sacra","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T5","locator":"Sacra estimates Turing reached $350M in annualized revenue at the end of 2025, about 17% above the end of 2024 (about $300M); page gives no 2026 figure"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Method: Sacra (T4) estimates US$350M annualised at the end of 2025 (+17% on about US$300M); the company said in January 2025 that it had reached a US$300M ARR (press release, T2). Sacra's page gives no 2026 figure. Turing's own homepage lists datasets, RL environments and an expert network for labs, and Sacra notes 1,000+ enterprise clients of Turing Intelligence. Central US$350M for the in-scope part, low US$200M (part of the revenue is enterprise AI work), high US$600M with 2026 growth. Split follows the card notes. Main input: Sacra (T4). Share = company ÷ function total; low and high move only this company's own range and hold the others at central. Turing independent model evaluation: US$7m (2-18m) = run-rate US$350m (200-600m) × 2% (the smaller parts of a mixed vendor are varied by ±50% and the largest part absorbs the difference). Function total, annualised US$m (central): 586, made of identified candidates 566 and an allowance of 20 for vendors outside the register (for example Deccan AI, Shaip, Cogito Tech, Sigma AI, Centaur Labs, Revelo, e2f, Patronus AI, Haize Labs, Lightly, Mostly AI, Abundant, Ethos, Datumo and smaller environment builders). Largest, US$m: arena-ai 110, surge-ai 90, mercor 66, scale-ai 45, innodata 33, other candidates 222, vendors outside the register 20. Full table: a local evidence capture. Share 1.195% (range 0.344-3.016%). Source published date: undated (undated)."},"data-tooling":{"participates":false,"evidence":{"url":"https://www.turing.com/","title":"Turing homepage","publisher":"Turing","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T2","locator":"Offerings: off-the-shelf datasets and RL environments for frontier labs, expert network, enterprise AI; no labelling and data software offering described"}}},"activityGrowth":{"current":{"low":350000000,"central":350000000,"high":350000000,"unit":"USD","basis":"Annualised revenue run-rate in US dollars (latest month × 12 or latest quarter × 4)","period":"End of 2025","periodEnd":"2025-12-31","source":{"url":"https://sacra.com/c/turing/","title":"Turing revenue, funding and news","publisher":"Sacra","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T4","locator":"Sacra estimates Turing reached $350M in annualized revenue at the end of 2025, about 17% above the end of 2024 (about $300M); page gives no 2026 figure"},"note":"Sacra estimate for the end of 2025 (T4); includes enterprise work of Turing Intelligence."},"prior":{"low":300000000,"central":300000000,"high":300000000,"unit":"USD","basis":"Annualised revenue run-rate in US dollars (latest month × 12 or latest quarter × 4)","period":"End of 2024","periodEnd":"2024-12-31","source":{"url":"https://www.silicon.co.uk/press-release/turing-annual-revenue-runrate-nearly-triples-to-300m-in-profitable-year-as-company-sets-stage-for-agi-advancements-in-2025","title":"Turing Annual Revenue Runrate Nearly Triples to $300M in Profitable Year as Company Sets Stage for AGI Advancements in 2025","publisher":"Turing (press release via Business Wire and silicon.co.uk)","publishedDate":"2025-01-28","accessedDate":"2026-10-22","tier":"T2","locator":"The company says it reached a $300 million annual revenue run-rate (ARR), up from $100 million a year earlier, and became profitable in 2024"},"note":"The company says (press release, 28 Jan 2025) it reached a US$300M ARR in its profitable 2024; the release gives no exact date, so end of 2024 is an inference. Growth +17%; the window is the calendar year 2025, earlier than the other companies' windows, so no prior shares are given."}},"baselineRevenueUsd":{"low":350000000,"central":350000000,"high":350000000,"unit":"USD","basis":"Annualised revenue run-rate in US dollars (latest month × 12 or latest quarter × 4)","period":"End of 2025","periodEnd":"2025-12-31","source":{"url":"https://sacra.com/c/turing/","title":"Turing revenue, funding and news","publisher":"Sacra","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T4","locator":"Sacra estimates Turing reached $350M in annualized revenue at the end of 2025, about 17% above the end of 2024 (about $300M); page gives no 2026 figure"},"note":"Sacra estimate (T4); includes enterprise work of Turing Intelligence."},"screening":{"expert-data":{"levelLow":1,"levelHigh":3},"rl-environments":{"levelLow":1,"levelHigh":2},"evaluation":{"levelLow":1,"levelHigh":2}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"74afdb9bdce296ad3c06055e73ed9522a722bfa44cc021af0683bee7b59c63fc"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"74afdb9bdce296ad3c06055e73ed9522a722bfa44cc021af0683bee7b59c63fc"}}},"innodata":{"schema":"5lc-index-inputs/1","key":"models-data-evaluation","entityId":"innodata","name":"Innodata","kind":"public","commercialActivityVerified":true,"shares":{"expert-data":{"low":1.488,"central":2.795,"high":4.121,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"Q2 2026, annualised","periodEnd":"2026-06-30","source":{"url":"https://www.sec.gov/Archives/edgar/data/0000903651/000110465926092010/tm2621499d1_ex99-1.htm","title":"Innodata Reports Record Second Quarter 2026 Results (8-K exhibit 99.1)","publisher":"Innodata Inc. (SEC Form 8-K)","publishedDate":"2026-08-06","accessedDate":"2026-10-22","tier":"T5","locator":"Q2 2026 revenue $92.1M (six months $182.2M); largest customer 37% of revenue, a Big Tech customer 34%; growth 'increasingly research driven'"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Method: Q2 2026 revenue US$92.1M x 4 = US$368.6M (Form 8-K, T1). The largest customer was 37% of Q2 revenue and a Big Tech customer 34%, so most revenue is model-developer work, but some is enterprise data engineering, so central is taken at US$330M. Split follows the card notes (expert 60, labelling 30, evaluation 10). Full pass: Q2 2026 revenue was up 58% on Q2 2025 (US$58.4M); the company reiterated 2026 revenue growth guidance of 40% or more and did not report segment revenue. Main input: Innodata Inc. (SEC Form 8-K) (T1). Share = company ÷ function total; low and high move only this company's own range and hold the others at central. Innodata expert data for post-training: US$198m (104-296m) = run-rate US$330m (260-370m) × 60% (the smaller parts of a mixed vendor are varied by ±50% and the largest part absorbs the difference). Function total, annualised US$m (central): 7,084, made of identified candidates 6,934 and an allowance of 150 for vendors outside the register (for example Deccan AI, Shaip, Cogito Tech, Sigma AI, Centaur Labs, Revelo, e2f, Patronus AI, Haize Labs, Lightly, Mostly AI, Abundant, Ethos, Datumo and smaller environment builders). Largest, US$m: mercor 1,650, surge-ai 1,296, handshake-ai 882, micro1 850, scale-ai 412, other candidates 1,844, vendors outside the register 150. Full table: a local evidence capture. Share 2.795% (range 1.488-4.121%). Source published date: 2026-08-06 (full date)."},"labelling-multimodal":{"low":2.295,"central":5.627,"high":9.114,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"Q2 2026, annualised","periodEnd":"2026-06-30","source":{"url":"https://www.sec.gov/Archives/edgar/data/0000903651/000110465926092010/tm2621499d1_ex99-1.htm","title":"Innodata Reports Record Second Quarter 2026 Results (8-K exhibit 99.1)","publisher":"Innodata Inc. (SEC Form 8-K)","publishedDate":"2026-08-06","accessedDate":"2026-10-22","tier":"T5","locator":"Q2 2026 revenue $92.1M (six months $182.2M); largest customer 37% of revenue, a Big Tech customer 34%; growth 'increasingly research driven'"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Method: Q2 2026 revenue US$92.1M x 4 = US$368.6M (Form 8-K, T1). The largest customer was 37% of Q2 revenue and a Big Tech customer 34%, so most revenue is model-developer work, but some is enterprise data engineering, so central is taken at US$330M. Split follows the card notes (expert 60, labelling 30, evaluation 10). Full pass: Q2 2026 revenue was up 58% on Q2 2025 (US$58.4M); the company reiterated 2026 revenue growth guidance of 40% or more and did not report segment revenue. Main input: Innodata Inc. (SEC Form 8-K) (T1). Share = company ÷ function total; low and high move only this company's own range and hold the others at central. Innodata large-scale labelling and multimodal data: US$99m (39-166m) = run-rate US$330m (260-370m) × 30% (the smaller parts of a mixed vendor are varied by ±50% and the largest part absorbs the difference). Function total, annualised US$m (central): 1,759, made of identified candidates 1,609 and an allowance of 150 for vendors outside the register (for example Deccan AI, Shaip, Cogito Tech, Sigma AI, Centaur Labs, Revelo, e2f, Patronus AI, Haize Labs, Lightly, Mostly AI, Abundant, Ethos, Datumo and smaller environment builders). Largest, US$m: scale-ai 262, appen 176, taskus 146, rws-trainai 140, telus-digital 124, other candidates 762, vendors outside the register 150. Full table: a local evidence capture. Share 5.627% (range 2.295-9.114%). Source published date: 2026-08-06 (full date)."},"rl-environments":{"participates":false,"evidence":{"url":"https://www.sec.gov/Archives/edgar/data/0000903651/000110465926092010/tm2621499d1_ex99-1.htm","title":"Innodata Reports Record Second Quarter 2026 Results (8-K exhibit 99.1)","publisher":"Innodata Inc. (SEC Form 8-K)","publishedDate":"2026-08-06","accessedDate":"2026-10-22","tier":"T1","locator":"Q2 2026 revenue $92.1M (six months $182.2M); largest customer 37% of revenue, a Big Tech customer 34%; growth 'increasingly research driven'; no RL environments and agent tasks offering described"}},"evaluation":{"low":2.297,"central":5.633,"high":9.123,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"Q2 2026, annualised","periodEnd":"2026-06-30","source":{"url":"https://www.sec.gov/Archives/edgar/data/0000903651/000110465926092010/tm2621499d1_ex99-1.htm","title":"Innodata Reports Record Second Quarter 2026 Results (8-K exhibit 99.1)","publisher":"Innodata Inc. (SEC Form 8-K)","publishedDate":"2026-08-06","accessedDate":"2026-10-22","tier":"T5","locator":"Q2 2026 revenue $92.1M (six months $182.2M); largest customer 37% of revenue, a Big Tech customer 34%; growth 'increasingly research driven'"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Method: Q2 2026 revenue US$92.1M x 4 = US$368.6M (Form 8-K, T1). The largest customer was 37% of Q2 revenue and a Big Tech customer 34%, so most revenue is model-developer work, but some is enterprise data engineering, so central is taken at US$330M. Split follows the card notes (expert 60, labelling 30, evaluation 10). Full pass: Q2 2026 revenue was up 58% on Q2 2025 (US$58.4M); the company reiterated 2026 revenue growth guidance of 40% or more and did not report segment revenue. Main input: Innodata Inc. (SEC Form 8-K) (T1). Share = company ÷ function total; low and high move only this company's own range and hold the others at central. Innodata independent model evaluation: US$33m (13-55.5m) = run-rate US$330m (260-370m) × 10% (the smaller parts of a mixed vendor are varied by ±50% and the largest part absorbs the difference). Function total, annualised US$m (central): 586, made of identified candidates 566 and an allowance of 20 for vendors outside the register (for example Deccan AI, Shaip, Cogito Tech, Sigma AI, Centaur Labs, Revelo, e2f, Patronus AI, Haize Labs, Lightly, Mostly AI, Abundant, Ethos, Datumo and smaller environment builders). Largest, US$m: arena-ai 110, surge-ai 90, mercor 66, scale-ai 45, innodata 33, other candidates 222, vendors outside the register 20. Full table: a local evidence capture. Share 5.633% (range 2.297-9.123%). Source published date: 2026-08-06 (full date)."},"data-tooling":{"participates":false,"evidence":{"url":"https://www.sec.gov/Archives/edgar/data/0000903651/000110465926092010/tm2621499d1_ex99-1.htm","title":"Innodata Reports Record Second Quarter 2026 Results (8-K exhibit 99.1)","publisher":"Innodata Inc. (SEC Form 8-K)","publishedDate":"2026-08-06","accessedDate":"2026-10-22","tier":"T1","locator":"Q2 2026 revenue $92.1M (six months $182.2M); largest customer 37% of revenue, a Big Tech customer 34%; growth 'increasingly research driven'; no labelling and data software offering described"}}},"priorShares":{"expert-data":{"low":2.254,"central":4.206,"high":6.16,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"Q2 2025 on the current basis (Q3 2025 frame)","periodEnd":"2025-06-30","source":{"url":"https://www.sec.gov/Archives/edgar/data/0000903651/000110465926092010/tm2621499d1_ex99-1.htm","title":"Innodata Reports Record Second Quarter 2026 Results (8-K exhibit 99.1)","publisher":"Innodata Inc. (SEC Form 8-K)","publishedDate":"2026-08-06","accessedDate":"2026-10-22","tier":"T5","locator":"Q2 2026 revenue $92.1M (six months $182.2M); largest customer 37% of revenue, a Big Tech customer 34%; growth 'increasingly research driven'"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Prior share in the Q3 2025 frame. Q2 2025 revenue was US$58.4M (x 4 = US$233.6M) against US$368.6M now (x 1.578), so the prior is today's in-scope central US$330M ÷ 1.578 = US$209.1M. Innodata expert data for post-training: US$125m (65.9-188m) = prior run-rate US$209m (165-234m) × 60% (split and the ±50% variation of the smaller parts held as today). Prior part total US$2,983m = the sum over all vendors of Q3 2025 run-rate × split, plus the allowance of 150 ÷ 1.5 = 100. Q3 2025 frame: each vendor's run-rate at the time, documented where it exists (Mercor US$500M, Handshake AI US$80M, micro1 US$45M, Snorkel US$20.8M, Arena and Proximal none, Appen, TaskUs and Innodata from their H1 or Q2 2025 reports scaled to today's basis, Surge US$1.2bn, Scale's data business US$850M, Turing US$337.5M, Fleet US$1M, Labelbox one fifth of today's, RWS today's divided by 1.45), and every other vendor at today's run-rate divided by 1.5 (the rounded median, 1.54, of the 12 documented multiples); each vendor's split across the five parts is held at today's. The frame is an estimate and its uncertainty is not in the share ranges. Share 4.206% (range 2.254-6.16%). The current share is 2.795% on the same basis, so the change is -1.4 points."},"labelling-multimodal":{"low":1.894,"central":4.672,"high":7.615,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"Q2 2025 on the current basis (Q3 2025 frame)","periodEnd":"2025-06-30","source":{"url":"https://www.sec.gov/Archives/edgar/data/0000903651/000110465926092010/tm2621499d1_ex99-1.htm","title":"Innodata Reports Record Second Quarter 2026 Results (8-K exhibit 99.1)","publisher":"Innodata Inc. (SEC Form 8-K)","publishedDate":"2026-08-06","accessedDate":"2026-10-22","tier":"T5","locator":"Q2 2026 revenue $92.1M (six months $182.2M); largest customer 37% of revenue, a Big Tech customer 34%; growth 'increasingly research driven'"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Prior share in the Q3 2025 frame. Q2 2025 revenue was US$58.4M (x 4 = US$233.6M) against US$368.6M now (x 1.578), so the prior is today's in-scope central US$330M ÷ 1.578 = US$209.1M. Innodata large-scale labelling and multimodal data: US$62.7m (24.7-106m) = prior run-rate US$209m (165-234m) × 30% (split and the ±50% variation of the smaller parts held as today). Prior part total US$1,343m = the sum over all vendors of Q3 2025 run-rate × split, plus the allowance of 150 ÷ 1.5 = 100. Q3 2025 frame: each vendor's run-rate at the time, documented where it exists (Mercor US$500M, Handshake AI US$80M, micro1 US$45M, Snorkel US$20.8M, Arena and Proximal none, Appen, TaskUs and Innodata from their H1 or Q2 2025 reports scaled to today's basis, Surge US$1.2bn, Scale's data business US$850M, Turing US$337.5M, Fleet US$1M, Labelbox one fifth of today's, RWS today's divided by 1.45), and every other vendor at today's run-rate divided by 1.5 (the rounded median, 1.54, of the 12 documented multiples); each vendor's split across the five parts is held at today's. The frame is an estimate and its uncertainty is not in the share ranges. Share 4.672% (range 1.894-7.615%). The current share is 5.627% on the same basis, so the change is +1.0 points."},"evaluation":{"low":3.265,"central":7.891,"high":12.594,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"Q2 2025 on the current basis (Q3 2025 frame)","periodEnd":"2025-06-30","source":{"url":"https://www.sec.gov/Archives/edgar/data/0000903651/000110465926092010/tm2621499d1_ex99-1.htm","title":"Innodata Reports Record Second Quarter 2026 Results (8-K exhibit 99.1)","publisher":"Innodata Inc. (SEC Form 8-K)","publishedDate":"2026-08-06","accessedDate":"2026-10-22","tier":"T5","locator":"Q2 2026 revenue $92.1M (six months $182.2M); largest customer 37% of revenue, a Big Tech customer 34%; growth 'increasingly research driven'"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Prior share in the Q3 2025 frame. Q2 2025 revenue was US$58.4M (x 4 = US$233.6M) against US$368.6M now (x 1.578), so the prior is today's in-scope central US$330M ÷ 1.578 = US$209.1M. Innodata independent model evaluation: US$20.9m (8.24-35.2m) = prior run-rate US$209m (165-234m) × 10% (split and the ±50% variation of the smaller parts held as today). Prior part total US$265m = the sum over all vendors of Q3 2025 run-rate × split, plus the allowance of 20 ÷ 1.5 = 13.3. Q3 2025 frame: each vendor's run-rate at the time, documented where it exists (Mercor US$500M, Handshake AI US$80M, micro1 US$45M, Snorkel US$20.8M, Arena and Proximal none, Appen, TaskUs and Innodata from their H1 or Q2 2025 reports scaled to today's basis, Surge US$1.2bn, Scale's data business US$850M, Turing US$337.5M, Fleet US$1M, Labelbox one fifth of today's, RWS today's divided by 1.45), and every other vendor at today's run-rate divided by 1.5 (the rounded median, 1.54, of the 12 documented multiples); each vendor's split across the five parts is held at today's. The frame is an estimate and its uncertainty is not in the share ranges. Share 7.891% (range 3.265-12.594%). The current share is 5.633% on the same basis, so the change is -2.3 points."}},"activityGrowth":{"current":{"low":368568000,"central":368568000,"high":368568000,"unit":"USD","basis":"Quarterly total revenue × 4 in US dollars (includes some enterprise work outside this sublayer)","period":"Q2 2026, annualised","periodEnd":"2026-06-30","source":{"url":"https://www.sec.gov/Archives/edgar/data/0000903651/000110465926092010/tm2621499d1_ex99-1.htm","title":"Innodata Reports Record Second Quarter 2026 Results (8-K exhibit 99.1)","publisher":"Innodata Inc. (SEC Form 8-K)","publishedDate":"2026-08-06","accessedDate":"2026-10-22","tier":"T1","locator":"Q2 2026 revenue $92.1M (six months $182.2M); largest customer 37% of revenue, a Big Tech customer 34%; growth 'increasingly research driven'"},"note":"Q2 2026 revenue US$92.142M × 4 (Form 8-K exhibit, 6 Aug 2026)."},"prior":{"low":233572000,"central":233572000,"high":233572000,"unit":"USD","basis":"Quarterly total revenue × 4 in US dollars (includes some enterprise work outside this sublayer)","period":"Q2 2025, annualised","periodEnd":"2025-06-30","source":{"url":"https://www.sec.gov/Archives/edgar/data/0000903651/000110465926092010/tm2621499d1_ex99-1.htm","title":"Innodata Reports Record Second Quarter 2026 Results (8-K exhibit 99.1)","publisher":"Innodata Inc. (SEC Form 8-K)","publishedDate":"2026-08-06","accessedDate":"2026-10-22","tier":"T1","locator":"Q2 2026 revenue $92.1M (six months $182.2M); largest customer 37% of revenue, a Big Tech customer 34%; growth 'increasingly research driven'"},"note":"Q2 2025 revenue US$58.393M × 4 (comparative column of the same exhibit). Growth +57.8%."}},"baselineRevenueUsd":{"low":368568000,"central":368568000,"high":368568000,"unit":"USD","basis":"Quarterly total revenue × 4 in US dollars (includes some enterprise work outside this sublayer)","period":"Q2 2026, annualised","periodEnd":"2026-06-30","offBasis":true,"source":{"url":"https://www.sec.gov/Archives/edgar/data/0000903651/000110465926092010/tm2621499d1_ex99-1.htm","title":"Innodata Reports Record Second Quarter 2026 Results (8-K exhibit 99.1)","publisher":"Innodata Inc. (SEC Form 8-K)","publishedDate":"2026-08-06","accessedDate":"2026-10-22","tier":"T1","locator":"Q2 2026 revenue $92.1M (six months $182.2M); largest customer 37% of revenue, a Big Tech customer 34%; growth 'increasingly research driven'"},"note":"Q2 2026 revenue US$92.142M × 4 (Form 8-K exhibit). Annual revenue for 2025 was not used so that every company is on a current basis."},"screening":{"expert-data":{"levelLow":1,"levelHigh":3},"labelling-multimodal":{"levelLow":1,"levelHigh":3},"evaluation":{"levelLow":1,"levelHigh":2}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"630318f3b22ede9f9bbc22001e2224c44db6d81ad18ec783e618f6f8deb74363"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"630318f3b22ede9f9bbc22001e2224c44db6d81ad18ec783e618f6f8deb74363"}}},"prolific":{"schema":"5lc-index-inputs/1","key":"models-data-evaluation","entityId":"prolific","name":"Prolific","kind":"private","commercialActivityVerified":true,"shares":{"expert-data":{"low":1.342,"central":2.964,"high":5.822,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"Annualised gross revenue, April 2026","periodEnd":"2026-04-30","source":{"url":"https://sacra.com/research/350m-year-mercor-for-human-personality/","title":"$350M/year Mercor for human personality","publisher":"Sacra","publishedDate":"2026-05-19","accessedDate":"2026-10-22","tier":"T5","locator":"Sacra estimates Prolific reached roughly $350M in annualized revenue in April 2026, including payments to contractors; customers are frontier AI labs, AI app developers and enterprises"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Method: Sacra (T4, 19 May 2026; https://sacra.com/research/350m-year-mercor-for-human-personality/, opened) estimates about US$350M annualised in April 2026 including contractor pay (gross); its page gives no earlier figure and no growth rate. Customers are frontier labs, AI app developers and enterprises. The range is a 5LayerCake judgment (T5) built on that one figure and on no other source: central US$300M in scope (about 85% of US$350M, because part of the revenue is academic and market research); low US$170M (about half of US$350M, if the research share is larger than assumed); high US$500M = US$350M x 1.43, an uplift of about 6% a month over the six months from April to the October 2026 run date (well below the roughly doubling in a few months documented for Mercor, Handshake AI and micro1 in this sublayer), with the whole amount counted in scope. A Business Post headline seen in search results (15 May 2026) points higher; that page was refused (HTTP 403), has not been opened and is not relied on. Split follows the card notes (Toloka and Prolific). Full pass: no second source found. Main input: Sacra (T4). Share = company ÷ function total; low and high move only this company's own range and hold the others at central. Prolific expert data for post-training: US$210m (93.5-425m) = run-rate US$300m (170-500m) × 70% (the smaller parts of a mixed vendor are varied by ±50% and the largest part absorbs the difference). Function total, annualised US$m (central): 7,084, made of identified candidates 6,934 and an allowance of 150 for vendors outside the register (for example Deccan AI, Shaip, Cogito Tech, Sigma AI, Centaur Labs, Revelo, e2f, Patronus AI, Haize Labs, Lightly, Mostly AI, Abundant, Ethos, Datumo and smaller environment builders). Largest, US$m: mercor 1,650, surge-ai 1,296, handshake-ai 882, micro1 850, scale-ai 412, other candidates 1,844, vendors outside the register 150. Full table: a local evidence capture. Share 2.964% (range 1.342-5.822%). Source published date: 2026-05-19 (full date)."},"labelling-multimodal":{"low":0.991,"central":3.411,"high":8.111,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"Annualised gross revenue, April 2026","periodEnd":"2026-04-30","source":{"url":"https://sacra.com/research/350m-year-mercor-for-human-personality/","title":"$350M/year Mercor for human personality","publisher":"Sacra","publishedDate":"2026-05-19","accessedDate":"2026-10-22","tier":"T5","locator":"Sacra estimates Prolific reached roughly $350M in annualized revenue in April 2026, including payments to contractors; customers are frontier AI labs, AI app developers and enterprises"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Method: Sacra (T4, 19 May 2026; https://sacra.com/research/350m-year-mercor-for-human-personality/, opened) estimates about US$350M annualised in April 2026 including contractor pay (gross); its page gives no earlier figure and no growth rate. Customers are frontier labs, AI app developers and enterprises. The range is a 5LayerCake judgment (T5) built on that one figure and on no other source: central US$300M in scope (about 85% of US$350M, because part of the revenue is academic and market research); low US$170M (about half of US$350M, if the research share is larger than assumed); high US$500M = US$350M x 1.43, an uplift of about 6% a month over the six months from April to the October 2026 run date (well below the roughly doubling in a few months documented for Mercor, Handshake AI and micro1 in this sublayer), with the whole amount counted in scope. A Business Post headline seen in search results (15 May 2026) points higher; that page was refused (HTTP 403), has not been opened and is not relied on. Split follows the card notes (Toloka and Prolific). Full pass: no second source found. Main input: Sacra (T4). Share = company ÷ function total; low and high move only this company's own range and hold the others at central. Prolific large-scale labelling and multimodal data: US$60m (17-150m) = run-rate US$300m (170-500m) × 20% (the smaller parts of a mixed vendor are varied by ±50% and the largest part absorbs the difference). Function total, annualised US$m (central): 1,759, made of identified candidates 1,609 and an allowance of 150 for vendors outside the register (for example Deccan AI, Shaip, Cogito Tech, Sigma AI, Centaur Labs, Revelo, e2f, Patronus AI, Haize Labs, Lightly, Mostly AI, Abundant, Ethos, Datumo and smaller environment builders). Largest, US$m: scale-ai 262, appen 176, taskus 146, rws-trainai 140, telus-digital 124, other candidates 762, vendors outside the register 150. Full table: a local evidence capture. Share 3.411% (range 0.991-8.111%). Source published date: 2026-05-19 (full date)."},"rl-environments":{"participates":false,"evidence":{"url":"https://sacra.com/research/350m-year-mercor-for-human-personality/","title":"$350M/year Mercor for human personality","publisher":"Sacra","publishedDate":"2026-05-19","accessedDate":"2026-10-22","tier":"T4","locator":"Sacra estimates Prolific reached roughly $350M in annualized revenue in April 2026, including payments to contractors; customers are frontier AI labs, AI app developers and enterprises; no RL environments and agent tasks offering described"}},"evaluation":{"low":1.506,"central":5.121,"high":11.888,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"Annualised gross revenue, April 2026","periodEnd":"2026-04-30","source":{"url":"https://sacra.com/research/350m-year-mercor-for-human-personality/","title":"$350M/year Mercor for human personality","publisher":"Sacra","publishedDate":"2026-05-19","accessedDate":"2026-10-22","tier":"T5","locator":"Sacra estimates Prolific reached roughly $350M in annualized revenue in April 2026, including payments to contractors; customers are frontier AI labs, AI app developers and enterprises"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Method: Sacra (T4, 19 May 2026; https://sacra.com/research/350m-year-mercor-for-human-personality/, opened) estimates about US$350M annualised in April 2026 including contractor pay (gross); its page gives no earlier figure and no growth rate. Customers are frontier labs, AI app developers and enterprises. The range is a 5LayerCake judgment (T5) built on that one figure and on no other source: central US$300M in scope (about 85% of US$350M, because part of the revenue is academic and market research); low US$170M (about half of US$350M, if the research share is larger than assumed); high US$500M = US$350M x 1.43, an uplift of about 6% a month over the six months from April to the October 2026 run date (well below the roughly doubling in a few months documented for Mercor, Handshake AI and micro1 in this sublayer), with the whole amount counted in scope. A Business Post headline seen in search results (15 May 2026) points higher; that page was refused (HTTP 403), has not been opened and is not relied on. Split follows the card notes (Toloka and Prolific). Full pass: no second source found. Main input: Sacra (T4). Share = company ÷ function total; low and high move only this company's own range and hold the others at central. Prolific independent model evaluation: US$30m (8.5-75m) = run-rate US$300m (170-500m) × 10% (the smaller parts of a mixed vendor are varied by ±50% and the largest part absorbs the difference). Function total, annualised US$m (central): 586, made of identified candidates 566 and an allowance of 20 for vendors outside the register (for example Deccan AI, Shaip, Cogito Tech, Sigma AI, Centaur Labs, Revelo, e2f, Patronus AI, Haize Labs, Lightly, Mostly AI, Abundant, Ethos, Datumo and smaller environment builders). Largest, US$m: arena-ai 110, surge-ai 90, mercor 66, scale-ai 45, innodata 33, other candidates 222, vendors outside the register 20. Full table: a local evidence capture. Share 5.121% (range 1.506-11.888%). Source published date: 2026-05-19 (full date)."},"data-tooling":{"participates":false,"evidence":{"url":"https://sacra.com/research/350m-year-mercor-for-human-personality/","title":"$350M/year Mercor for human personality","publisher":"Sacra","publishedDate":"2026-05-19","accessedDate":"2026-10-22","tier":"T4","locator":"Sacra estimates Prolific reached roughly $350M in annualized revenue in April 2026, including payments to contractors; customers are frontier AI labs, AI app developers and enterprises; no labelling and data software offering described"}}},"baselineRevenueUsd":{"low":350000000,"central":350000000,"high":350000000,"unit":"USD","basis":"Annualised gross revenue in US dollars (includes payments to contractors and non-AI research work)","period":"April 2026 run-rate","periodEnd":"2026-04-30","offBasis":true,"source":{"url":"https://sacra.com/research/350m-year-mercor-for-human-personality/","title":"$350M/year Mercor for human personality","publisher":"Sacra","publishedDate":"2026-05-19","accessedDate":"2026-10-22","tier":"T4","locator":"Sacra estimates Prolific reached roughly $350M in annualized revenue in April 2026, including payments to contractors; customers are frontier AI labs, AI app developers and enterprises"},"note":"Sacra estimate (T4). Only part of this is data for model developers (the Reach figures use about US$300M), so it is flagged off basis."},"screening":{"expert-data":{"levelLow":1,"levelHigh":3},"labelling-multimodal":{"levelLow":1,"levelHigh":2},"evaluation":{"levelLow":1,"levelHigh":2}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"22e21c382caed254734633f74e9ae3494267fad887e6fe4d1ff7dfaa56199a16"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"22e21c382caed254734633f74e9ae3494267fad887e6fe4d1ff7dfaa56199a16"}}},"appen":{"schema":"5lc-index-inputs/1","key":"models-data-evaluation","entityId":"appen","name":"Appen","kind":"public","commercialActivityVerified":true,"shares":{"expert-data":{"low":0.597,"central":1.334,"high":2.204,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"June 2026 annualised run-rate (China) plus Q2 2026 annualised (Global)","periodEnd":"2026-06-30","source":{"url":"https://announcements.asx.com.au/asxpdf/20260827/pdf/0739d4q7lyh3w2.pdf","title":"H1 FY26 results","publisher":"Appen Limited (ASX announcement)","publishedDate":"2026-08-27","accessedDate":"2026-10-22","tier":"T5","locator":"H1 FY26 revenue $119.9M (China $76.2M, annualised run-rate above $175M in June; Global $43.7M, Q2 +20% on Q1); FY26 guidance $270-300M"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Method: ASX announcement (27 Aug 2026): Appen China's annualised run-rate exceeded US$175M in June; Appen Global H1 revenue US$43.7M with Q2 20% above Q1 (Q2 about US$23.8M, annualised about US$95M). Total about US$270M, matching FY26 guidance of US$270-300M; low US$240M is H1 x 2. Split follows the card notes (Appen China is mostly generative-AI and prebuilt-dataset work). Full pass: H1 FY26 revenue was US$119.9M, up 17% on H1 FY25; Appen China US$76.2M (+80%), Appen Global US$43.7M (-27%); the China run-rate was US$135M at the end of 2025. Main input: Appen Limited (ASX announcement) (T1). Share = company ÷ function total; low and high move only this company's own range and hold the others at central. Appen expert data for post-training: US$94.5m (42-157m) = run-rate US$270m (240-300m) × 35% (the smaller parts of a mixed vendor are varied by ±50% and the largest part absorbs the difference). Function total, annualised US$m (central): 7,084, made of identified candidates 6,934 and an allowance of 150 for vendors outside the register (for example Deccan AI, Shaip, Cogito Tech, Sigma AI, Centaur Labs, Revelo, e2f, Patronus AI, Haize Labs, Lightly, Mostly AI, Abundant, Ethos, Datumo and smaller environment builders). Largest, US$m: mercor 1,650, surge-ai 1,296, handshake-ai 882, micro1 850, scale-ai 412, other candidates 1,844, vendors outside the register 150. Full table: a local evidence capture. Share 1.334% (range 0.597-2.204%). Source published date: 2026-08-27 (full date)."},"labelling-multimodal":{"low":6.715,"central":9.976,"high":13.515,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"June 2026 annualised run-rate (China) plus Q2 2026 annualised (Global)","periodEnd":"2026-06-30","source":{"url":"https://announcements.asx.com.au/asxpdf/20260827/pdf/0739d4q7lyh3w2.pdf","title":"H1 FY26 results","publisher":"Appen Limited (ASX announcement)","publishedDate":"2026-08-27","accessedDate":"2026-10-22","tier":"T5","locator":"H1 FY26 revenue $119.9M (China $76.2M, annualised run-rate above $175M in June; Global $43.7M, Q2 +20% on Q1); FY26 guidance $270-300M"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Method: ASX announcement (27 Aug 2026): Appen China's annualised run-rate exceeded US$175M in June; Appen Global H1 revenue US$43.7M with Q2 20% above Q1 (Q2 about US$23.8M, annualised about US$95M). Total about US$270M, matching FY26 guidance of US$270-300M; low US$240M is H1 x 2. Split follows the card notes (Appen China is mostly generative-AI and prebuilt-dataset work). Full pass: H1 FY26 revenue was US$119.9M, up 17% on H1 FY25; Appen China US$76.2M (+80%), Appen Global US$43.7M (-27%); the China run-rate was US$135M at the end of 2025. Main input: Appen Limited (ASX announcement) (T1). Share = company ÷ function total; low and high move only this company's own range and hold the others at central. Appen large-scale labelling and multimodal data: US$176m (114-248m) = run-rate US$270m (240-300m) × 65% (the smaller parts of a mixed vendor are varied by ±50% and the largest part absorbs the difference). Function total, annualised US$m (central): 1,759, made of identified candidates 1,609 and an allowance of 150 for vendors outside the register (for example Deccan AI, Shaip, Cogito Tech, Sigma AI, Centaur Labs, Revelo, e2f, Patronus AI, Haize Labs, Lightly, Mostly AI, Abundant, Ethos, Datumo and smaller environment builders). Largest, US$m: scale-ai 262, appen 176, taskus 146, rws-trainai 140, telus-digital 124, other candidates 762, vendors outside the register 150. Full table: a local evidence capture. Share 9.976% (range 6.715-13.515%). Source published date: 2026-08-27 (full date)."},"rl-environments":{"participates":false,"evidence":{"url":"https://announcements.asx.com.au/asxpdf/20260827/pdf/0739d4q7lyh3w2.pdf","title":"H1 FY26 results","publisher":"Appen Limited (ASX announcement)","publishedDate":"2026-08-27","accessedDate":"2026-10-22","tier":"T1","locator":"H1 FY26 revenue $119.9M (China $76.2M, annualised run-rate above $175M in June; Global $43.7M, Q2 +20% on Q1); FY26 guidance $270-300M; no RL environments and agent tasks offering described"}},"evaluation":{"participates":false,"evidence":{"url":"https://announcements.asx.com.au/asxpdf/20260827/pdf/0739d4q7lyh3w2.pdf","title":"H1 FY26 results","publisher":"Appen Limited (ASX announcement)","publishedDate":"2026-08-27","accessedDate":"2026-10-22","tier":"T1","locator":"H1 FY26 revenue $119.9M (China $76.2M, annualised run-rate above $175M in June; Global $43.7M, Q2 +20% on Q1); FY26 guidance $270-300M; no independent model evaluation offering described"}},"data-tooling":{"participates":false,"evidence":{"url":"https://announcements.asx.com.au/asxpdf/20260827/pdf/0739d4q7lyh3w2.pdf","title":"H1 FY26 results","publisher":"Appen Limited (ASX announcement)","publishedDate":"2026-08-27","accessedDate":"2026-10-22","tier":"T1","locator":"H1 FY26 revenue $119.9M (China $76.2M, annualised run-rate above $175M in June; Global $43.7M, Q2 +20% on Q1); FY26 guidance $270-300M; no labelling and data software offering described"}}},"priorShares":{"expert-data":{"low":1.367,"central":2.708,"high":4.024,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"H1 FY25 on the current basis (Q3 2025 frame)","periodEnd":"2025-06-30","source":{"url":"https://announcements.asx.com.au/asxpdf/20260827/pdf/0739d4q7lyh3w2.pdf","title":"H1 FY26 results","publisher":"Appen Limited (ASX announcement)","publishedDate":"2026-08-27","accessedDate":"2026-10-22","tier":"T5","locator":"H1 FY26 revenue $119.9M (China $76.2M, annualised run-rate above $175M in June; Global $43.7M, Q2 +20% on Q1); FY26 guidance $270-300M"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Prior share in the Q3 2025 frame. H1 FY25 revenue was US$102.5M (= US$119.9M ÷ 1.17; growth of +17% is rounded), × 2 = US$205M on the half-year basis, against US$239.8M now; applied to today's central US$270M that gives US$230.8M (US$229.8-232.3M). Appen expert data for post-training: US$80.8m (40.2-122m) = prior run-rate US$231m (230-232m) × 35% (split and the ±50% variation of the smaller parts held as today). Prior part total US$2,983m = the sum over all vendors of Q3 2025 run-rate × split, plus the allowance of 150 ÷ 1.5 = 100. Q3 2025 frame: each vendor's run-rate at the time, documented where it exists (Mercor US$500M, Handshake AI US$80M, micro1 US$45M, Snorkel US$20.8M, Arena and Proximal none, Appen, TaskUs and Innodata from their H1 or Q2 2025 reports scaled to today's basis, Surge US$1.2bn, Scale's data business US$850M, Turing US$337.5M, Fleet US$1M, Labelbox one fifth of today's, RWS today's divided by 1.45), and every other vendor at today's run-rate divided by 1.5 (the rounded median, 1.54, of the 12 documented multiples); each vendor's split across the five parts is held at today's. The frame is an estimate and its uncertainty is not in the share ranges. Share 2.708% (range 1.367-4.024%). The current share is 1.334% on the same basis, so the change is -1.4 points."},"labelling-multimodal":{"low":8.383,"central":11.17,"high":13.814,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"H1 FY25 on the current basis (Q3 2025 frame)","periodEnd":"2025-06-30","source":{"url":"https://announcements.asx.com.au/asxpdf/20260827/pdf/0739d4q7lyh3w2.pdf","title":"H1 FY26 results","publisher":"Appen Limited (ASX announcement)","publishedDate":"2026-08-27","accessedDate":"2026-10-22","tier":"T5","locator":"H1 FY26 revenue $119.9M (China $76.2M, annualised run-rate above $175M in June; Global $43.7M, Q2 +20% on Q1); FY26 guidance $270-300M"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Prior share in the Q3 2025 frame. H1 FY25 revenue was US$102.5M (= US$119.9M ÷ 1.17; growth of +17% is rounded), × 2 = US$205M on the half-year basis, against US$239.8M now; applied to today's central US$270M that gives US$230.8M (US$229.8-232.3M). Appen large-scale labelling and multimodal data: US$150m (109-191m) = prior run-rate US$231m (230-232m) × 65% (split and the ±50% variation of the smaller parts held as today). Prior part total US$1,343m = the sum over all vendors of Q3 2025 run-rate × split, plus the allowance of 150 ÷ 1.5 = 100. Q3 2025 frame: each vendor's run-rate at the time, documented where it exists (Mercor US$500M, Handshake AI US$80M, micro1 US$45M, Snorkel US$20.8M, Arena and Proximal none, Appen, TaskUs and Innodata from their H1 or Q2 2025 reports scaled to today's basis, Surge US$1.2bn, Scale's data business US$850M, Turing US$337.5M, Fleet US$1M, Labelbox one fifth of today's, RWS today's divided by 1.45), and every other vendor at today's run-rate divided by 1.5 (the rounded median, 1.54, of the 12 documented multiples); each vendor's split across the five parts is held at today's. The frame is an estimate and its uncertainty is not in the share ranges. Share 11.17% (range 8.383-13.814%). The current share is 9.976% on the same basis, so the change is -1.2 points."}},"activityGrowth":{"current":{"low":239800000,"central":239800000,"high":239800000,"unit":"USD","basis":"Half-year revenue × 2 in US dollars (all Appen revenue)","period":"H1 FY26, annualised","periodEnd":"2026-06-30","source":{"url":"https://announcements.asx.com.au/asxpdf/20260827/pdf/0739d4q7lyh3w2.pdf","title":"H1 FY26 results","publisher":"Appen Limited (ASX announcement)","publishedDate":"2026-08-27","accessedDate":"2026-10-22","tier":"T1","locator":"H1 FY26 revenue $119.9M (China $76.2M, annualised run-rate above $175M in June; Global $43.7M, Q2 +20% on Q1); FY26 guidance $270-300M"},"note":"H1 FY26 revenue US$119.9M × 2 (ASX announcement, 27 Aug 2026)."},"prior":{"low":204100000,"central":205000000,"high":205800000,"unit":"USD","basis":"Half-year revenue × 2 in US dollars (all Appen revenue)","period":"H1 FY25, annualised","periodEnd":"2025-06-30","source":{"url":"https://announcements.asx.com.au/asxpdf/20260827/pdf/0739d4q7lyh3w2.pdf","title":"H1 FY26 results","publisher":"Appen Limited (ASX announcement)","publishedDate":"2026-08-27","accessedDate":"2026-10-22","tier":"T1","locator":"H1 FY26 revenue $119.9M (China $76.2M, annualised run-rate above $175M in June; Global $43.7M, Q2 +20% on Q1); FY26 guidance $270-300M"},"note":"Derived: H1 FY26 revenue was +17% against the prior corresponding period, so H1 FY25 was US$119.9M ÷ 1.17 = US$102.5M (US$102.0-102.9M if +17% is rounded), × 2 = US$205M. Growth +17%."}},"baselineRevenueUsd":{"low":239800000,"central":239800000,"high":239800000,"unit":"USD","basis":"Half-year revenue × 2 in US dollars (all Appen revenue)","period":"H1 FY26, annualised","periodEnd":"2026-06-30","source":{"url":"https://announcements.asx.com.au/asxpdf/20260827/pdf/0739d4q7lyh3w2.pdf","title":"H1 FY26 results","publisher":"Appen Limited (ASX announcement)","publishedDate":"2026-08-27","accessedDate":"2026-10-22","tier":"T1","locator":"H1 FY26 revenue $119.9M (China $76.2M, annualised run-rate above $175M in June; Global $43.7M, Q2 +20% on Q1); FY26 guidance $270-300M"},"note":"H1 FY26 revenue US$119.9M × 2 (ASX announcement, 27 Aug 2026); FY26 guidance is US$270-300M."},"screening":{"expert-data":{"levelLow":1,"levelHigh":3},"labelling-multimodal":{"levelLow":1,"levelHigh":3}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"c27db37f3647f02fd0e57001a069628abc525cca25ec193b9788bc2dbbb99f69"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"c27db37f3647f02fd0e57001a069628abc525cca25ec193b9788bc2dbbb99f69"}}},"taskus":{"schema":"5lc-index-inputs/1","key":"models-data-evaluation","entityId":"taskus","name":"TaskUs","kind":"public","commercialActivityVerified":true,"shares":{"expert-data":{"low":0.694,"central":1.445,"high":2.391,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"Half-year to June 2026, annualised (x 2)","periodEnd":"2026-06-30","source":{"url":"https://csv-storage.forexpros.com/slides/c7ad004a2be7f7acb476cfc22640135f657a661c0d33df530e0bb4da3a481d97.pdf","title":"TaskUs Investor Presentation - Q2 2026","publisher":"TaskUs (deck hosted by Investing.com)","publishedDate":"2026-08-05","accessedDate":"2026-10-22","tier":"T5","locator":"Slide 30 (revenue reconciliation by service offering): AI Services $128,044K for the six months to June 2026 against $98,023K a year earlier, $214,218K for 2025 and $244,239K for the twelve months to June 2026; slide 3: AI Services +25.8% year over year in Q2 2026"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Method: The company's Q2 2026 investor deck gives AI Services revenue of US$128.0M for the first half of 2026 (US$98.0M in H1 2025), US$244.2M for the twelve months to June 2026 and US$214.2M for 2025, with 25.8% growth in Q2 (8-K); half-year x 2 = US$256M. Central US$256M, low the twelve-month US$244M, high US$285M with continued growth. AI Services covers autonomous-vehicle and robotics data, generative-AI response writing and ranking, and AI content-moderation training; trust and safety and customer experience are excluded. Split follows the card notes. Independent evaluation (added at the check-fix pass of 22 October 2026): the deck says AI Services also provides 'expert response writing, ranking and scoring, prompt review, and adversarial testing' for generative-AI clients (slide 11) and lists 'Adversarial Red Teaming' among its AI-safety services (slide 16). No split is published, so 3% of the AI Services run-rate is taken from labelling (60% to 57%) as a judgment. AI-safety work sold in the Trust & Safety line (US$302M for the twelve months to June 2026) is not counted because the deck gives no figure for it. Main input: TaskUs (deck hosted by Investing.com) (T5). Share = company ÷ function total; low and high move only this company's own range and hold the others at central. TaskUs expert data for post-training: US$102m (48.8-171m) = run-rate US$256m (244-285m) × 40% (the smaller parts of a mixed vendor are varied by ±50% and the largest part absorbs the difference). Function total, annualised US$m (central): 7,084, made of identified candidates 6,934 and an allowance of 150 for vendors outside the register (for example Deccan AI, Shaip, Cogito Tech, Sigma AI, Centaur Labs, Revelo, e2f, Patronus AI, Haize Labs, Lightly, Mostly AI, Abundant, Ethos, Datumo and smaller environment builders). Largest, US$m: mercor 1,650, surge-ai 1,296, handshake-ai 882, micro1 850, scale-ai 412, other candidates 1,844, vendors outside the register 150. Full table: a local evidence capture. Share 1.445% (range 0.694-2.391%). Source published date: 2026-08-05 (full date)."},"labelling-multimodal":{"low":5.095,"central":8.294,"high":12.178,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"Half-year to June 2026, annualised (x 2)","periodEnd":"2026-06-30","source":{"url":"https://csv-storage.forexpros.com/slides/c7ad004a2be7f7acb476cfc22640135f657a661c0d33df530e0bb4da3a481d97.pdf","title":"TaskUs Investor Presentation - Q2 2026","publisher":"TaskUs (deck hosted by Investing.com)","publishedDate":"2026-08-05","accessedDate":"2026-10-22","tier":"T5","locator":"Slide 30 (revenue reconciliation by service offering): AI Services $128,044K for the six months to June 2026 against $98,023K a year earlier, $214,218K for 2025 and $244,239K for the twelve months to June 2026; slide 3: AI Services +25.8% year over year in Q2 2026"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Method: The company's Q2 2026 investor deck gives AI Services revenue of US$128.0M for the first half of 2026 (US$98.0M in H1 2025), US$244.2M for the twelve months to June 2026 and US$214.2M for 2025, with 25.8% growth in Q2 (8-K); half-year x 2 = US$256M. Central US$256M, low the twelve-month US$244M, high US$285M with continued growth. AI Services covers autonomous-vehicle and robotics data, generative-AI response writing and ranking, and AI content-moderation training; trust and safety and customer experience are excluded. Split follows the card notes. Independent evaluation (added at the check-fix pass of 22 October 2026): the deck says AI Services also provides 'expert response writing, ranking and scoring, prompt review, and adversarial testing' for generative-AI clients (slide 11) and lists 'Adversarial Red Teaming' among its AI-safety services (slide 16). No split is published, so 3% of the AI Services run-rate is taken from labelling (60% to 57%) as a judgment. AI-safety work sold in the Trust & Safety line (US$302M for the twelve months to June 2026) is not counted because the deck gives no figure for it. Main input: TaskUs (deck hosted by Investing.com) (T5). Share = company ÷ function total; low and high move only this company's own range and hold the others at central. TaskUs large-scale labelling and multimodal data: US$146m (86.6-224m) = run-rate US$256m (244-285m) × 57% (the smaller parts of a mixed vendor are varied by ±50% and the largest part absorbs the difference). Function total, annualised US$m (central): 1,759, made of identified candidates 1,609 and an allowance of 150 for vendors outside the register (for example Deccan AI, Shaip, Cogito Tech, Sigma AI, Centaur Labs, Revelo, e2f, Patronus AI, Haize Labs, Lightly, Mostly AI, Abundant, Ethos, Datumo and smaller environment builders). Largest, US$m: scale-ai 262, appen 176, taskus 146, rws-trainai 140, telus-digital 124, other candidates 762, vendors outside the register 150. Full table: a local evidence capture. Share 8.294% (range 5.095-12.178%). Source published date: 2026-08-05 (full date)."},"rl-environments":{"participates":false,"evidence":{"url":"https://csv-storage.forexpros.com/slides/c7ad004a2be7f7acb476cfc22640135f657a661c0d33df530e0bb4da3a481d97.pdf","title":"TaskUs Investor Presentation - Q2 2026","publisher":"TaskUs (deck hosted by Investing.com)","publishedDate":"2026-08-05","accessedDate":"2026-10-22","tier":"T2","locator":"Slide 11: AI Services is 20% of TTM 6/26 revenue, $244mm (autonomous vehicles, generative AI, object identification, robotics, OCR); slide 3: AI Services +25.8% year over year in Q2 2026; no RL environments and agent tasks offering described"}},"evaluation":{"low":0.629,"central":1.311,"high":2.17,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"Half-year to June 2026, annualised (x 2)","periodEnd":"2026-06-30","source":{"url":"https://csv-storage.forexpros.com/slides/c7ad004a2be7f7acb476cfc22640135f657a661c0d33df530e0bb4da3a481d97.pdf","title":"TaskUs Investor Presentation - Q2 2026","publisher":"TaskUs (deck hosted by Investing.com)","publishedDate":"2026-08-05","accessedDate":"2026-10-22","tier":"T5","locator":"Slide 11: AI Services also provides 'expert response writing, ranking and scoring, prompt review, and adversarial testing for our GenAI clients'; slide 16: 'Adversarial Red Teaming'; slide 30 (revenue reconciliation by service offering): AI Services $128,044K for the six months to June 2026, $98,023K a year earlier, $244,239K for the twelve months to June 2026"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Method: The company's Q2 2026 investor deck gives AI Services revenue of US$128.0M for the first half of 2026 (US$98.0M in H1 2025), US$244.2M for the twelve months to June 2026 and US$214.2M for 2025, with 25.8% growth in Q2 (8-K); half-year x 2 = US$256M. Central US$256M, low the twelve-month US$244M, high US$285M with continued growth. AI Services covers autonomous-vehicle and robotics data, generative-AI response writing and ranking, and AI content-moderation training; trust and safety and customer experience are excluded. Split follows the card notes. Independent evaluation (added at the check-fix pass of 22 October 2026): the deck says AI Services also provides 'expert response writing, ranking and scoring, prompt review, and adversarial testing' for generative-AI clients (slide 11) and lists 'Adversarial Red Teaming' among its AI-safety services (slide 16). No split is published, so 3% of the AI Services run-rate is taken from labelling (60% to 57%) as a judgment. AI-safety work sold in the Trust & Safety line (US$302M for the twelve months to June 2026) is not counted because the deck gives no figure for it. Main input: TaskUs (deck hosted by Investing.com) (T5). Share = company ÷ function total; low and high move only this company's own range and hold the others at central. TaskUs independent model evaluation: US$7.68m (3.66-12.8m) = run-rate US$256m (244-285m) × 3% (the smaller parts of a mixed vendor are varied by ±50% and the largest part absorbs the difference). Function total, annualised US$m (central): 586, made of identified candidates 566 and an allowance of 20 for vendors outside the register (for example Deccan AI, Shaip, Cogito Tech, Sigma AI, Centaur Labs, Revelo, e2f, Patronus AI, Haize Labs, Lightly, Mostly AI, Abundant, Ethos, Datumo and smaller environment builders). Largest, US$m: arena-ai 110, surge-ai 90, mercor 66, scale-ai 45, innodata 33, other candidates 222, vendors outside the register 20. Full table: a local evidence capture. Share 1.311% (range 0.629-2.17%). Source published date: 2026-08-05 (full date)."},"data-tooling":{"participates":false,"evidence":{"url":"https://csv-storage.forexpros.com/slides/c7ad004a2be7f7acb476cfc22640135f657a661c0d33df530e0bb4da3a481d97.pdf","title":"TaskUs Investor Presentation - Q2 2026","publisher":"TaskUs (deck hosted by Investing.com)","publishedDate":"2026-08-05","accessedDate":"2026-10-22","tier":"T2","locator":"Slide 11: AI Services is 20% of TTM 6/26 revenue, $244mm (autonomous vehicles, generative AI, object identification, robotics, OCR); slide 3: AI Services +25.8% year over year in Q2 2026; no labelling and data software offering described"}}},"priorShares":{"expert-data":{"low":1.27,"central":2.628,"high":4.312,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"H1 2025 on the current basis (Q3 2025 frame)","periodEnd":"2025-06-30","source":{"url":"https://csv-storage.forexpros.com/slides/c7ad004a2be7f7acb476cfc22640135f657a661c0d33df530e0bb4da3a481d97.pdf","title":"TaskUs Investor Presentation - Q2 2026","publisher":"TaskUs (deck hosted by Investing.com)","publishedDate":"2026-08-05","accessedDate":"2026-10-22","tier":"T5","locator":"Slide 30 (revenue reconciliation by service offering): AI Services $128,044K for the six months to June 2026 against $98,023K a year earlier, $214,218K for 2025 and $244,239K for the twelve months to June 2026; slide 3: AI Services +25.8% year over year in Q2 2026"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Prior share in the Q3 2025 frame. AI Services revenue was US$98.0M in H1 2025 against US$128.0M in H1 2026 (x 1.306), so the prior is today's central US$256M ÷ 1.306 = US$196.0M. TaskUs expert data for post-training: US$78.4m (37.4-131m) = prior run-rate US$196m (187-218m) × 40% (split and the ±50% variation of the smaller parts held as today). Prior part total US$2,983m = the sum over all vendors of Q3 2025 run-rate × split, plus the allowance of 150 ÷ 1.5 = 100. Q3 2025 frame: each vendor's run-rate at the time, documented where it exists (Mercor US$500M, Handshake AI US$80M, micro1 US$45M, Snorkel US$20.8M, Arena and Proximal none, Appen, TaskUs and Innodata from their H1 or Q2 2025 reports scaled to today's basis, Surge US$1.2bn, Scale's data business US$850M, Turing US$337.5M, Fleet US$1M, Labelbox one fifth of today's, RWS today's divided by 1.45), and every other vendor at today's run-rate divided by 1.5 (the rounded median, 1.54, of the 12 documented multiples); each vendor's split across the five parts is held at today's. The frame is an estimate and its uncertainty is not in the share ranges. Share 2.628% (range 1.27-4.312%). The current share is 1.445% on the same basis, so the change is -1.2 points."},"labelling-multimodal":{"low":5.111,"central":8.318,"high":12.212,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"H1 2025 on the current basis (Q3 2025 frame)","periodEnd":"2025-06-30","source":{"url":"https://csv-storage.forexpros.com/slides/c7ad004a2be7f7acb476cfc22640135f657a661c0d33df530e0bb4da3a481d97.pdf","title":"TaskUs Investor Presentation - Q2 2026","publisher":"TaskUs (deck hosted by Investing.com)","publishedDate":"2026-08-05","accessedDate":"2026-10-22","tier":"T5","locator":"Slide 30 (revenue reconciliation by service offering): AI Services $128,044K for the six months to June 2026 against $98,023K a year earlier, $214,218K for 2025 and $244,239K for the twelve months to June 2026; slide 3: AI Services +25.8% year over year in Q2 2026"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Prior share in the Q3 2025 frame. AI Services revenue was US$98.0M in H1 2025 against US$128.0M in H1 2026 (x 1.306), so the prior is today's central US$256M ÷ 1.306 = US$196.0M. TaskUs large-scale labelling and multimodal data: US$112m (66.3-171m) = prior run-rate US$196m (187-218m) × 57% (split and the ±50% variation of the smaller parts held as today). Prior part total US$1,343m = the sum over all vendors of Q3 2025 run-rate × split, plus the allowance of 150 ÷ 1.5 = 100. Q3 2025 frame: each vendor's run-rate at the time, documented where it exists (Mercor US$500M, Handshake AI US$80M, micro1 US$45M, Snorkel US$20.8M, Arena and Proximal none, Appen, TaskUs and Innodata from their H1 or Q2 2025 reports scaled to today's basis, Surge US$1.2bn, Scale's data business US$850M, Turing US$337.5M, Fleet US$1M, Labelbox one fifth of today's, RWS today's divided by 1.45), and every other vendor at today's run-rate divided by 1.5 (the rounded median, 1.54, of the 12 documented multiples); each vendor's split across the five parts is held at today's. The frame is an estimate and its uncertainty is not in the share ranges. Share 8.318% (range 5.111-12.212%). The current share is 8.294% on the same basis, so the change is -0.0 points."},"evaluation":{"low":1.07,"central":2.218,"high":3.65,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"H1 2025 on the current basis (Q3 2025 frame)","periodEnd":"2025-06-30","source":{"url":"https://csv-storage.forexpros.com/slides/c7ad004a2be7f7acb476cfc22640135f657a661c0d33df530e0bb4da3a481d97.pdf","title":"TaskUs Investor Presentation - Q2 2026","publisher":"TaskUs (deck hosted by Investing.com)","publishedDate":"2026-08-05","accessedDate":"2026-10-22","tier":"T5","locator":"Slide 30 (revenue reconciliation by service offering): AI Services $128,044K for the six months to June 2026 against $98,023K a year earlier, $214,218K for 2025 and $244,239K for the twelve months to June 2026; slide 3: AI Services +25.8% year over year in Q2 2026"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Prior share in the Q3 2025 frame. AI Services revenue was US$98.0M in H1 2025 against US$128.0M in H1 2026 (x 1.306), so the prior is today's central US$256M ÷ 1.306 = US$196.0M. TaskUs independent model evaluation: US$5.88m (2.8-9.82m) = prior run-rate US$196m (187-218m) × 3% (split and the ±50% variation of the smaller parts held as today). Prior part total US$265m = the sum over all vendors of Q3 2025 run-rate × split, plus the allowance of 20 ÷ 1.5 = 13.3. Q3 2025 frame: each vendor's run-rate at the time, documented where it exists (Mercor US$500M, Handshake AI US$80M, micro1 US$45M, Snorkel US$20.8M, Arena and Proximal none, Appen, TaskUs and Innodata from their H1 or Q2 2025 reports scaled to today's basis, Surge US$1.2bn, Scale's data business US$850M, Turing US$337.5M, Fleet US$1M, Labelbox one fifth of today's, RWS today's divided by 1.45), and every other vendor at today's run-rate divided by 1.5 (the rounded median, 1.54, of the 12 documented multiples); each vendor's split across the five parts is held at today's. The frame is an estimate and its uncertainty is not in the share ranges. Share 2.218% (range 1.07-3.65%). The current share is 1.311% on the same basis, so the change is -0.9 points."}},"activityGrowth":{"current":{"low":256088000,"central":256088000,"high":256088000,"unit":"USD","basis":"Half-year AI Services revenue × 2 in US dollars","period":"H1 2026, annualised","periodEnd":"2026-06-30","source":{"url":"https://csv-storage.forexpros.com/slides/c7ad004a2be7f7acb476cfc22640135f657a661c0d33df530e0bb4da3a481d97.pdf","title":"TaskUs Investor Presentation - Q2 2026","publisher":"TaskUs (deck hosted by Investing.com)","publishedDate":"2026-08-05","accessedDate":"2026-10-22","tier":"T5","locator":"Slide 30 (revenue reconciliation by service offering): AI Services $128,044K for the six months to June 2026 against $98,023K a year earlier, $214,218K for 2025 and $244,239K for the twelve months to June 2026; slide 3: AI Services +25.8% year over year in Q2 2026"},"note":"AI Services revenue for the six months to June 2026 US$128.044M × 2 (revenue reconciliation by service offering, investor deck)."},"prior":{"low":196046000,"central":196046000,"high":196046000,"unit":"USD","basis":"Half-year AI Services revenue × 2 in US dollars","period":"H1 2025, annualised","periodEnd":"2025-06-30","source":{"url":"https://csv-storage.forexpros.com/slides/c7ad004a2be7f7acb476cfc22640135f657a661c0d33df530e0bb4da3a481d97.pdf","title":"TaskUs Investor Presentation - Q2 2026","publisher":"TaskUs (deck hosted by Investing.com)","publishedDate":"2026-08-05","accessedDate":"2026-10-22","tier":"T5","locator":"Slide 30 (revenue reconciliation by service offering): AI Services $128,044K for the six months to June 2026 against $98,023K a year earlier, $214,218K for 2025 and $244,239K for the twelve months to June 2026; slide 3: AI Services +25.8% year over year in Q2 2026"},"note":"AI Services revenue for the six months to June 2025 US$98.023M × 2 (same table). Growth +30.6%."}},"baselineRevenueUsd":{"low":244239000,"central":244239000,"high":244239000,"unit":"USD","basis":"AI Services revenue, twelve months to June 2026, US dollars","period":"Twelve months to June 2026","periodEnd":"2026-06-30","source":{"url":"https://csv-storage.forexpros.com/slides/c7ad004a2be7f7acb476cfc22640135f657a661c0d33df530e0bb4da3a481d97.pdf","title":"TaskUs Investor Presentation - Q2 2026","publisher":"TaskUs (deck hosted by Investing.com)","publishedDate":"2026-08-05","accessedDate":"2026-10-22","tier":"T5","locator":"Slide 30 (revenue reconciliation by service offering): AI Services $128,044K for the six months to June 2026 against $98,023K a year earlier, $214,218K for 2025 and $244,239K for the twelve months to June 2026; slide 3: AI Services +25.8% year over year in Q2 2026"},"note":"The investor deck's revenue reconciliation gives AI Services revenue of US$244,239K for the twelve months to June 2026 (the company says)."},"screening":{"expert-data":{"levelLow":1,"levelHigh":3},"labelling-multimodal":{"levelLow":1,"levelHigh":3},"evaluation":{"levelLow":1,"levelHigh":2}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"07b2954e76d94ed3f9435ed9113ddfc368a5674db676334d11559cdb5c26727c"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"07b2954e76d94ed3f9435ed9113ddfc368a5674db676334d11559cdb5c26727c"}}},"invisible-technologies":{"schema":"5lc-index-inputs/1","key":"models-data-evaluation","entityId":"invisible-technologies","name":"Invisible Technologies","kind":"private","commercialActivityVerified":true,"shares":{"expert-data":{"low":0.744,"central":2.117,"high":5.454,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"CY2024 revenue (T3), extrapolated to a 2026 run-rate (estimate)","periodEnd":"2024-12-31","source":{"url":"https://siliconangle.com/2025/09/16/ai-data-provider-invisible-raises-100m-2b-valuation/","title":"AI data provider Invisible raises $100M at $2B valuation","publisher":"SiliconANGLE","publishedDate":"2025-09-16","accessedDate":"2026-10-22","tier":"T5","locator":"Between 2020 and 2024, Invisible's annual revenue increased by a factor of over 48, to $134 million; the deal values it at more than $2 billion (Bloomberg)"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Method: Revenue was US$134M in 2024 (SiliconANGLE, Sep 2025, T3), when it raised US$100M at above US$2bn; no 2026 figure is published. Central US$250M applies roughly 1.9x for two years of sector growth; low US$130M is the 2024 level, high US$500M. Part of its revenue is enterprise software and process work, outside this sublayer. Its site lists AI training data and RL environments for labs (Cohere, ElevenLabs, Microsoft logos; 24,000+ experts). Split follows the card notes with 10% environments added from the site. Full pass: no 2025 or 2026 figure found; Sacra's page also ends at its US$134M estimate for 2024 (up from about US$60M in 2023). Main input: SiliconANGLE (T3). Share = company ÷ function total; low and high move only this company's own range and hold the others at central. Invisible Technologies expert data for post-training: US$150m (52-400m) = run-rate US$250m (130-500m) × 60% (the smaller parts of a mixed vendor are varied by ±50% and the largest part absorbs the difference). Function total, annualised US$m (central): 7,084, made of identified candidates 6,934 and an allowance of 150 for vendors outside the register (for example Deccan AI, Shaip, Cogito Tech, Sigma AI, Centaur Labs, Revelo, e2f, Patronus AI, Haize Labs, Lightly, Mostly AI, Abundant, Ethos, Datumo and smaller environment builders). Largest, US$m: mercor 1,650, surge-ai 1,296, handshake-ai 882, micro1 850, scale-ai 412, other candidates 1,844, vendors outside the register 150. Full table: a local evidence capture. Share 2.117% (range 0.744-5.454%). Second source(s): Invisible Technologies undated (T2): Offerings: AI training data and RL environments for labs, enterprise AI platform; named logos include Cohere, ElevenLabs, Microsoft; 24,000+ experts. Source published date: 2025-09-16 (full date)."},"labelling-multimodal":{"low":1.145,"central":4.263,"high":11.785,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"CY2024 revenue (T3), extrapolated to a 2026 run-rate (estimate)","periodEnd":"2024-12-31","source":{"url":"https://siliconangle.com/2025/09/16/ai-data-provider-invisible-raises-100m-2b-valuation/","title":"AI data provider Invisible raises $100M at $2B valuation","publisher":"SiliconANGLE","publishedDate":"2025-09-16","accessedDate":"2026-10-22","tier":"T5","locator":"Between 2020 and 2024, Invisible's annual revenue increased by a factor of over 48, to $134 million; the deal values it at more than $2 billion (Bloomberg)"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Method: Revenue was US$134M in 2024 (SiliconANGLE, Sep 2025, T3), when it raised US$100M at above US$2bn; no 2026 figure is published. Central US$250M applies roughly 1.9x for two years of sector growth; low US$130M is the 2024 level, high US$500M. Part of its revenue is enterprise software and process work, outside this sublayer. Its site lists AI training data and RL environments for labs (Cohere, ElevenLabs, Microsoft logos; 24,000+ experts). Split follows the card notes with 10% environments added from the site. Full pass: no 2025 or 2026 figure found; Sacra's page also ends at its US$134M estimate for 2024 (up from about US$60M in 2023). Main input: SiliconANGLE (T3). Share = company ÷ function total; low and high move only this company's own range and hold the others at central. Invisible Technologies large-scale labelling and multimodal data: US$75m (19.5-225m) = run-rate US$250m (130-500m) × 30% (the smaller parts of a mixed vendor are varied by ±50% and the largest part absorbs the difference). Function total, annualised US$m (central): 1,759, made of identified candidates 1,609 and an allowance of 150 for vendors outside the register (for example Deccan AI, Shaip, Cogito Tech, Sigma AI, Centaur Labs, Revelo, e2f, Patronus AI, Haize Labs, Lightly, Mostly AI, Abundant, Ethos, Datumo and smaller environment builders). Largest, US$m: scale-ai 262, appen 176, taskus 146, rws-trainai 140, telus-digital 124, other candidates 762, vendors outside the register 150. Full table: a local evidence capture. Share 4.263% (range 1.145-11.785%). Second source(s): Invisible Technologies undated (T2): Offerings: AI training data and RL environments for labs, enterprise AI platform; named logos include Cohere, ElevenLabs, Microsoft; 24,000+ experts. Source published date: 2025-09-16 (full date)."},"rl-environments":{"low":0.357,"central":1.36,"high":3.971,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"CY2024 revenue (T3), extrapolated to a 2026 run-rate (estimate)","periodEnd":"2024-12-31","source":{"url":"https://siliconangle.com/2025/09/16/ai-data-provider-invisible-raises-100m-2b-valuation/","title":"AI data provider Invisible raises $100M at $2B valuation","publisher":"SiliconANGLE","publishedDate":"2025-09-16","accessedDate":"2026-10-22","tier":"T5","locator":"Between 2020 and 2024, Invisible's annual revenue increased by a factor of over 48, to $134 million; the deal values it at more than $2 billion (Bloomberg)"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Method: Revenue was US$134M in 2024 (SiliconANGLE, Sep 2025, T3), when it raised US$100M at above US$2bn; no 2026 figure is published. Central US$250M applies roughly 1.9x for two years of sector growth; low US$130M is the 2024 level, high US$500M. Part of its revenue is enterprise software and process work, outside this sublayer. Its site lists AI training data and RL environments for labs (Cohere, ElevenLabs, Microsoft logos; 24,000+ experts). Split follows the card notes with 10% environments added from the site. Full pass: no 2025 or 2026 figure found; Sacra's page also ends at its US$134M estimate for 2024 (up from about US$60M in 2023). Main input: SiliconANGLE (T3). Share = company ÷ function total; low and high move only this company's own range and hold the others at central. Invisible Technologies RL environments and agent tasks: US$25m (6.5-75m) = run-rate US$250m (130-500m) × 10% (the smaller parts of a mixed vendor are varied by ±50% and the largest part absorbs the difference). Function total, annualised US$m (central): 1,839, made of identified candidates 1,739 and an allowance of 100 for vendors outside the register (for example Deccan AI, Shaip, Cogito Tech, Sigma AI, Centaur Labs, Revelo, e2f, Patronus AI, Haize Labs, Lightly, Mostly AI, Abundant, Ethos, Datumo and smaller environment builders). Largest, US$m: mercor 484, surge-ai 414, proximal 140, handshake-ai 136, micro1 130, other candidates 434, vendors outside the register 100. Full table: a local evidence capture. Share 1.36% (range 0.357-3.971%). Second source(s): Invisible Technologies undated (T2): Offerings: AI training data and RL environments for labs, enterprise AI platform; named logos include Cohere, ElevenLabs, Microsoft; 24,000+ experts. Source published date: 2025-09-16 (full date)."},"evaluation":{"participates":false,"evidence":{"url":"https://invisibletech.ai/","title":"Invisible homepage","publisher":"Invisible Technologies","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T2","locator":"Offerings: AI training data and RL environments for labs, enterprise AI platform; named logos include Cohere, ElevenLabs, Microsoft; 24,000+ experts; no independent model evaluation offering described"}},"data-tooling":{"participates":false,"evidence":{"url":"https://invisibletech.ai/","title":"Invisible homepage","publisher":"Invisible Technologies","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T2","locator":"Offerings: AI training data and RL environments for labs, enterprise AI platform; named logos include Cohere, ElevenLabs, Microsoft; 24,000+ experts; no labelling and data software offering described"}}},"baselineRevenueUsd":{"low":134000000,"central":134000000,"high":134000000,"unit":"USD","basis":"Annual revenue, calendar year, US dollars (total company revenue; includes enterprise work outside this sublayer)","period":"CY2024","periodEnd":"2024-12-31","offBasis":true,"source":{"url":"https://siliconangle.com/2025/09/16/ai-data-provider-invisible-raises-100m-2b-valuation/","title":"AI data provider Invisible raises $100M at $2B valuation","publisher":"SiliconANGLE","publishedDate":"2025-09-16","accessedDate":"2026-10-22","tier":"T3","locator":"Between 2020 and 2024, Invisible's annual revenue increased by a factor of over 48, to $134 million; the deal values it at more than $2 billion (Bloomberg)"},"note":"SiliconANGLE (T3): annual revenue of US$134M in 2024. No later figure found."},"screening":{"expert-data":{"levelLow":1,"levelHigh":3},"labelling-multimodal":{"levelLow":1,"levelHigh":2},"rl-environments":{"levelLow":1,"levelHigh":2}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"06a4514239564bf1ceda8cdcbdeb80ba40857bb75ead36736a9b2d48f9083a42"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"06a4514239564bf1ceda8cdcbdeb80ba40857bb75ead36736a9b2d48f9083a42"}}},"telus-digital":{"schema":"5lc-index-inputs/1","key":"models-data-evaluation","entityId":"telus-digital","name":"TELUS Digital","kind":"unit","parentId":"telus","commercialActivityVerified":true,"shares":{"expert-data":{"low":0.305,"central":1.163,"high":2.855,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"Q2 2026 evidence (ramp-downs), carried to a 2026 estimate","periodEnd":"2026-06-30","source":{"url":"https://www.sec.gov/Archives/edgar/data/0000868675/000110465926088975/tm2615221d2_ex99-1.htm","title":"TELUS reports second quarter 2026 financial and operational results and resets quarterly dividend to support deleveraging and fuel long-term growth","publisher":"TELUS Corporation (SEC Form 6-K)","publishedDate":"2026-07-31","accessedDate":"2026-10-22","tier":"T5","locator":"Second quarter 2026, TELUS Digital segment (tables headed \"C$ millions\"): operating revenues C$653M against C$728M a year earlier (-10%), partly from \"client ramp-downs in our AI and data solutions service line\"; non-cash impairment of $2.1 billion relating to TELUS Digital (C$2,135M in the reconciliation); no AI-data dollar line"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Method: TELUS Digital does not report AI data revenue separately, and TELUS reports in Canadian dollars (tables headed 'C$ millions'). Inputs, all opened on 22 Oct 2026: (1) TELUS Q4 2025 results, Form 6-K, 12 Feb 2026 (https://www.sec.gov/Archives/edgar/data/868675/000110465926013856/tm261121d3_ex99-1.htm): the CEO says AI-enabling capabilities generated 'approximately $800 million in revenue in 2025' across TELUS Digital and TELUS Business Solutions; the sentence names no currency and the release reports in C$, so it is read as C$800M, which at the release's stated 2025 average rate of C$1.40 per US$1.00 is US$571M (800 / 1.40). The same release gives C$1.37 per US$1.00 as its 2026 average-rate assumption. The AI-enabling figure covers more than AI data (AI-enabled services and TELUS Business Solutions). (2) TELUS Q2 2026 results, Form 6-K, 31 Jul 2026 (T1; https://www.sec.gov/Archives/edgar/data/0000868675/000110465926088975/tm2615221d2_ex99-1.htm): TELUS Digital operating revenues C$653M against C$728M a year earlier (restated), down 10%, partly from 'client ramp-downs in our AI and data solutions service line', and a non-cash impairment of C$2.1bn on TELUS Digital (C$2,135M in the reconciliation); at C$1.37 the quarter is US$477M and the impairment about US$1.6bn. (3) TELUS Digital's 2024 results release (silicon.co.uk, https://www.silicon.co.uk/press-release/telus-digital-reports-fourth-quarter-and-full-year-2024-results-delivering-stable-operating-and-financial-performance-provides-outlook-for-2025) says only that AI Data Solutions showed 'good momentum' and gives no figure. Judgment (T5): AI data for model developers is about 40% of the US$571M AI-enabling total (range 25-60%; no split is published), which is US$229M for 2025 (US$143-343M), less 10% (range 0-25%) for the ramp-downs reported in 2026. Central 571 x 0.40 x 0.90 = US$206M, low 571 x 0.25 x 0.75 = US$107M, high 571 x 0.60 x 1.00 = US$343M. Restated at the check-fix pass of 22 October 2026: the earlier central of US$300M (150-450) had read the C$ figures as US dollars. Unit estimate only, not the parent's total. Split follows the card notes. Main input: TELUS Corporation (SEC Form 6-K) (T1). Share = company ÷ function total; low and high move only this company's own range and hold the others at central. TELUS Digital expert data for post-training: US$82.4m (21.4-206m) = run-rate US$206m (107-343m) × 40% (the smaller parts of a mixed vendor are varied by ±50% and the largest part absorbs the difference). Function total, annualised US$m (central): 7,084, made of identified candidates 6,934 and an allowance of 150 for vendors outside the register (for example Deccan AI, Shaip, Cogito Tech, Sigma AI, Centaur Labs, Revelo, e2f, Patronus AI, Haize Labs, Lightly, Mostly AI, Abundant, Ethos, Datumo and smaller environment builders). Largest, US$m: mercor 1,650, surge-ai 1,296, handshake-ai 882, micro1 850, scale-ai 412, other candidates 1,844, vendors outside the register 150. Full table: a local evidence capture. Share 1.163% (range 0.305-2.855%). Source published date: 2026-07-31 (full date)."},"labelling-multimodal":{"low":2.55,"central":7.026,"high":14.366,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"Q2 2026 evidence (ramp-downs), carried to a 2026 estimate","periodEnd":"2026-06-30","source":{"url":"https://www.sec.gov/Archives/edgar/data/0000868675/000110465926088975/tm2615221d2_ex99-1.htm","title":"TELUS reports second quarter 2026 financial and operational results and resets quarterly dividend to support deleveraging and fuel long-term growth","publisher":"TELUS Corporation (SEC Form 6-K)","publishedDate":"2026-07-31","accessedDate":"2026-10-22","tier":"T5","locator":"Second quarter 2026, TELUS Digital segment (tables headed \"C$ millions\"): operating revenues C$653M against C$728M a year earlier (-10%), partly from \"client ramp-downs in our AI and data solutions service line\"; non-cash impairment of $2.1 billion relating to TELUS Digital (C$2,135M in the reconciliation); no AI-data dollar line"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Method: TELUS Digital does not report AI data revenue separately, and TELUS reports in Canadian dollars (tables headed 'C$ millions'). Inputs, all opened on 22 Oct 2026: (1) TELUS Q4 2025 results, Form 6-K, 12 Feb 2026 (https://www.sec.gov/Archives/edgar/data/868675/000110465926013856/tm261121d3_ex99-1.htm): the CEO says AI-enabling capabilities generated 'approximately $800 million in revenue in 2025' across TELUS Digital and TELUS Business Solutions; the sentence names no currency and the release reports in C$, so it is read as C$800M, which at the release's stated 2025 average rate of C$1.40 per US$1.00 is US$571M (800 / 1.40). The same release gives C$1.37 per US$1.00 as its 2026 average-rate assumption. The AI-enabling figure covers more than AI data (AI-enabled services and TELUS Business Solutions). (2) TELUS Q2 2026 results, Form 6-K, 31 Jul 2026 (T1; https://www.sec.gov/Archives/edgar/data/0000868675/000110465926088975/tm2615221d2_ex99-1.htm): TELUS Digital operating revenues C$653M against C$728M a year earlier (restated), down 10%, partly from 'client ramp-downs in our AI and data solutions service line', and a non-cash impairment of C$2.1bn on TELUS Digital (C$2,135M in the reconciliation); at C$1.37 the quarter is US$477M and the impairment about US$1.6bn. (3) TELUS Digital's 2024 results release (silicon.co.uk, https://www.silicon.co.uk/press-release/telus-digital-reports-fourth-quarter-and-full-year-2024-results-delivering-stable-operating-and-financial-performance-provides-outlook-for-2025) says only that AI Data Solutions showed 'good momentum' and gives no figure. Judgment (T5): AI data for model developers is about 40% of the US$571M AI-enabling total (range 25-60%; no split is published), which is US$229M for 2025 (US$143-343M), less 10% (range 0-25%) for the ramp-downs reported in 2026. Central 571 x 0.40 x 0.90 = US$206M, low 571 x 0.25 x 0.75 = US$107M, high 571 x 0.60 x 1.00 = US$343M. Restated at the check-fix pass of 22 October 2026: the earlier central of US$300M (150-450) had read the C$ figures as US dollars. Unit estimate only, not the parent's total. Split follows the card notes. Main input: TELUS Corporation (SEC Form 6-K) (T1). Share = company ÷ function total; low and high move only this company's own range and hold the others at central. TELUS Digital large-scale labelling and multimodal data: US$124m (42.8-274m) = run-rate US$206m (107-343m) × 60% (the smaller parts of a mixed vendor are varied by ±50% and the largest part absorbs the difference). Function total, annualised US$m (central): 1,759, made of identified candidates 1,609 and an allowance of 150 for vendors outside the register (for example Deccan AI, Shaip, Cogito Tech, Sigma AI, Centaur Labs, Revelo, e2f, Patronus AI, Haize Labs, Lightly, Mostly AI, Abundant, Ethos, Datumo and smaller environment builders). Largest, US$m: scale-ai 262, appen 176, taskus 146, rws-trainai 140, telus-digital 124, other candidates 762, vendors outside the register 150. Full table: a local evidence capture. Share 7.026% (range 2.55-14.366%). Source published date: 2026-07-31 (full date)."},"rl-environments":{"participates":false,"evidence":{"url":"https://www.sec.gov/Archives/edgar/data/868675/000110465926013856/tm261121d3_ex99-1.htm","title":"TELUS Q4 2025 results (Form 6-K exhibit 99.1)","publisher":"TELUS Corporation (SEC Form 6-K)","publishedDate":"2026-02-12","accessedDate":"2026-10-22","tier":"T1","locator":"TELUS Digital Q4 2025 operating revenues C$722M (+2%; tables headed \"C$ millions\"); the CEO says AI-enabling capabilities generated approximately $800 million in revenue in 2025 across TELUS Digital and TELUS Business Solutions (no currency named in that sentence; read as C$ because the release reports in Canadian dollars); no AI-data line; no RL environments and agent tasks offering described"}},"evaluation":{"participates":false,"evidence":{"url":"https://www.sec.gov/Archives/edgar/data/868675/000110465926013856/tm261121d3_ex99-1.htm","title":"TELUS Q4 2025 results (Form 6-K exhibit 99.1)","publisher":"TELUS Corporation (SEC Form 6-K)","publishedDate":"2026-02-12","accessedDate":"2026-10-22","tier":"T1","locator":"TELUS Digital Q4 2025 operating revenues C$722M (+2%; tables headed \"C$ millions\"); the CEO says AI-enabling capabilities generated approximately $800 million in revenue in 2025 across TELUS Digital and TELUS Business Solutions (no currency named in that sentence; read as C$ because the release reports in Canadian dollars); no AI-data line; no independent model evaluation offering described"}},"data-tooling":{"participates":false,"evidence":{"url":"https://www.sec.gov/Archives/edgar/data/868675/000110465926013856/tm261121d3_ex99-1.htm","title":"TELUS Q4 2025 results (Form 6-K exhibit 99.1)","publisher":"TELUS Corporation (SEC Form 6-K)","publishedDate":"2026-02-12","accessedDate":"2026-10-22","tier":"T1","locator":"TELUS Digital Q4 2025 operating revenues C$722M (+2%; tables headed \"C$ millions\"); the CEO says AI-enabling capabilities generated approximately $800 million in revenue in 2025 across TELUS Digital and TELUS Business Solutions (no currency named in that sentence; read as C$ because the release reports in Canadian dollars); no AI-data line; no labelling and data software offering described"}}},"screening":{"expert-data":{"levelLow":1,"levelHigh":3},"labelling-multimodal":{"levelLow":1,"levelHigh":3}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"5f3d7feed8752429bcb185e9b2c0c0fdf59ef875f669d322cfb50d08cceba91e"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"5f3d7feed8752429bcb185e9b2c0c0fdf59ef875f669d322cfb50d08cceba91e"}}},"rws-trainai":{"schema":"5lc-index-inputs/1","key":"models-data-evaluation","entityId":"rws-trainai","name":"RWS TrainAI","kind":"unit","parentId":"rws","commercialActivityVerified":true,"shares":{"expert-data":{"low":0.256,"central":0.847,"high":1.762,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"Half-year to 31 March 2026, annualised","periodEnd":"2026-03-31","source":{"url":"https://www.investegate.co.uk/announcement/rns/rws-holdings--rws/half-year-financial-report/9612324","title":"RWS Holdings plc half-year financial report (six months to 31 March 2026)","publisher":"RWS Holdings plc (RNS)","publishedDate":"2026-06-11","accessedDate":"2026-10-22","tier":"T5","locator":"Generate segment (TrainAI and Content Technology) revenue GBP 99.4m (HY25 GBP 68.7m), growth mainly TrainAI; AI-related revenue 32% of group revenue GBP 360.3m; TrainAI not disclosed separately"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Method: RWS reports its Generate segment (TrainAI plus Content Technology) at GBP 99.4m for the six months to 31 March 2026 (HY25 GBP 68.7m), with growth mainly from TrainAI; TrainAI alone is not disclosed. Assume TrainAI is about 75% (55-90%) of Generate: GBP 74.6m x 2 = GBP 149m, at US$1.33 per GBP about US$200M (low US$120M, high US$280M). Unit estimate only (RWS group revenue GBP 360.3m for the half is not used). Full pass: the Generate segment grew 44.7% (52.0% organic constant-currency) in the half; the report credits an additional programme with an existing client and a new global technology client in TrainAI. Main input: RWS Holdings plc (RNS) (T1). Share = company ÷ function total; low and high move only this company's own range and hold the others at central. RWS TrainAI expert data for post-training: US$60m (18-126m) = run-rate US$200m (120-280m) × 30% (the smaller parts of a mixed vendor are varied by ±50% and the largest part absorbs the difference). Function total, annualised US$m (central): 7,084, made of identified candidates 6,934 and an allowance of 150 for vendors outside the register (for example Deccan AI, Shaip, Cogito Tech, Sigma AI, Centaur Labs, Revelo, e2f, Patronus AI, Haize Labs, Lightly, Mostly AI, Abundant, Ethos, Datumo and smaller environment builders). Largest, US$m: mercor 1,650, surge-ai 1,296, handshake-ai 882, micro1 850, scale-ai 412, other candidates 1,844, vendors outside the register 150. Full table: a local evidence capture. Share 0.847% (range 0.256-1.762%). Exchange rate assumed US$1.33 per GBP (not sourced; within the range of 2026 rates seen in the filings and covered by the stated range) Source published date: 2026-06-11 (full date)."},"labelling-multimodal":{"low":3.916,"central":7.958,"high":12.815,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"Half-year to 31 March 2026, annualised","periodEnd":"2026-03-31","source":{"url":"https://www.investegate.co.uk/announcement/rns/rws-holdings--rws/half-year-financial-report/9612324","title":"RWS Holdings plc half-year financial report (six months to 31 March 2026)","publisher":"RWS Holdings plc (RNS)","publishedDate":"2026-06-11","accessedDate":"2026-10-22","tier":"T5","locator":"Generate segment (TrainAI and Content Technology) revenue GBP 99.4m (HY25 GBP 68.7m), growth mainly TrainAI; AI-related revenue 32% of group revenue GBP 360.3m; TrainAI not disclosed separately"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Method: RWS reports its Generate segment (TrainAI plus Content Technology) at GBP 99.4m for the six months to 31 March 2026 (HY25 GBP 68.7m), with growth mainly from TrainAI; TrainAI alone is not disclosed. Assume TrainAI is about 75% (55-90%) of Generate: GBP 74.6m x 2 = GBP 149m, at US$1.33 per GBP about US$200M (low US$120M, high US$280M). Unit estimate only (RWS group revenue GBP 360.3m for the half is not used). Full pass: the Generate segment grew 44.7% (52.0% organic constant-currency) in the half; the report credits an additional programme with an existing client and a new global technology client in TrainAI. Main input: RWS Holdings plc (RNS) (T1). Share = company ÷ function total; low and high move only this company's own range and hold the others at central. RWS TrainAI large-scale labelling and multimodal data: US$140m (66-238m) = run-rate US$200m (120-280m) × 70% (the smaller parts of a mixed vendor are varied by ±50% and the largest part absorbs the difference). Function total, annualised US$m (central): 1,759, made of identified candidates 1,609 and an allowance of 150 for vendors outside the register (for example Deccan AI, Shaip, Cogito Tech, Sigma AI, Centaur Labs, Revelo, e2f, Patronus AI, Haize Labs, Lightly, Mostly AI, Abundant, Ethos, Datumo and smaller environment builders). Largest, US$m: scale-ai 262, appen 176, taskus 146, rws-trainai 140, telus-digital 124, other candidates 762, vendors outside the register 150. Full table: a local evidence capture. Share 7.958% (range 3.916-12.815%). Exchange rate assumed US$1.33 per GBP (not sourced; within the range of 2026 rates seen in the filings and covered by the stated range) Source published date: 2026-06-11 (full date)."},"rl-environments":{"participates":false,"evidence":{"url":"https://www.investegate.co.uk/announcement/rns/rws-holdings--rws/half-year-financial-report/9612324","title":"RWS Holdings plc half-year financial report (six months to 31 March 2026)","publisher":"RWS Holdings plc (RNS)","publishedDate":"2026-06-11","accessedDate":"2026-10-22","tier":"T1","locator":"Generate segment (TrainAI and Content Technology) revenue GBP 99.4m (HY25 GBP 68.7m), growth mainly TrainAI; AI-related revenue 32% of group revenue GBP 360.3m; TrainAI not disclosed separately; no RL environments and agent tasks offering described"}},"evaluation":{"participates":false,"evidence":{"url":"https://www.investegate.co.uk/announcement/rns/rws-holdings--rws/half-year-financial-report/9612324","title":"RWS Holdings plc half-year financial report (six months to 31 March 2026)","publisher":"RWS Holdings plc (RNS)","publishedDate":"2026-06-11","accessedDate":"2026-10-22","tier":"T1","locator":"Generate segment (TrainAI and Content Technology) revenue GBP 99.4m (HY25 GBP 68.7m), growth mainly TrainAI; AI-related revenue 32% of group revenue GBP 360.3m; TrainAI not disclosed separately; no independent model evaluation offering described"}},"data-tooling":{"participates":false,"evidence":{"url":"https://www.investegate.co.uk/announcement/rns/rws-holdings--rws/half-year-financial-report/9612324","title":"RWS Holdings plc half-year financial report (six months to 31 March 2026)","publisher":"RWS Holdings plc (RNS)","publishedDate":"2026-06-11","accessedDate":"2026-10-22","tier":"T1","locator":"Generate segment (TrainAI and Content Technology) revenue GBP 99.4m (HY25 GBP 68.7m), growth mainly TrainAI; AI-related revenue 32% of group revenue GBP 360.3m; TrainAI not disclosed separately; no labelling and data software offering described"}}},"activityGrowth":{"current":{"low":99400000,"central":99400000,"high":99400000,"unit":"GBP","basis":"Generate segment revenue (TrainAI and Content Technology), six months to 31 March, pounds sterling","period":"Half-year to 31 March 2026","periodEnd":"2026-03-31","source":{"url":"https://www.investegate.co.uk/announcement/rns/rws-holdings--rws/half-year-financial-report/9612324","title":"RWS Holdings plc half-year financial report (six months to 31 March 2026)","publisher":"RWS Holdings plc (RNS)","publishedDate":"2026-06-11","accessedDate":"2026-10-22","tier":"T1","locator":"Generate segment (TrainAI and Content Technology) revenue GBP 99.4m (HY25 GBP 68.7m), growth mainly TrainAI; AI-related revenue 32% of group revenue GBP 360.3m; TrainAI not disclosed separately"},"note":"Generate segment revenue GBP 99.4m (half-year report, 11 Jun 2026). A proxy: the segment includes Content Technology, which is outside this sublayer, while the report credits TrainAI with the growth."},"prior":{"low":68700000,"central":68700000,"high":68700000,"unit":"GBP","basis":"Generate segment revenue (TrainAI and Content Technology), six months to 31 March, pounds sterling","period":"Half-year to 31 March 2025","periodEnd":"2025-03-31","source":{"url":"https://www.investegate.co.uk/announcement/rns/rws-holdings--rws/half-year-financial-report/9612324","title":"RWS Holdings plc half-year financial report (six months to 31 March 2026)","publisher":"RWS Holdings plc (RNS)","publishedDate":"2026-06-11","accessedDate":"2026-10-22","tier":"T1","locator":"Generate segment (TrainAI and Content Technology) revenue GBP 99.4m (HY25 GBP 68.7m), growth mainly TrainAI; AI-related revenue 32% of group revenue GBP 360.3m; TrainAI not disclosed separately"},"note":"Generate segment revenue GBP 68.7m for HY25 (comparative in the same report). Growth +44.7% (52.0% organic constant-currency)."}},"screening":{"expert-data":{"levelLow":1,"levelHigh":2},"labelling-multimodal":{"levelLow":1,"levelHigh":3}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"36a06c739cfd1c45daeed44efc10e876c679a782c80793fffa566d5205b0ada7"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"36a06c739cfd1c45daeed44efc10e876c679a782c80793fffa566d5205b0ada7"}}},"proximal":{"schema":"5lc-index-inputs/1","key":"models-data-evaluation","entityId":"proximal","name":"Proximal","kind":"private","commercialActivityVerified":true,"shares":{"expert-data":{"low":0.256,"central":0.847,"high":1.576,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"Annualised run-rate, Q3 2026 (company claim)","periodEnd":"2026-09-30","source":{"url":"https://startupfortune.com/proximal-reveals-200-million-revenue-run-rate-and-15-million-seed-round/","title":"Proximal reveals $200 million revenue run rate and $15 million seed round","publisher":"Startup Fortune","publishedDate":"2026-10-09","accessedDate":"2026-10-22","tier":"T5","locator":"Proximal reports $200 million in annualized revenue within roughly ten months of operating; came out of stealth on September 29; customers are the labs building coding agents, none named (self-reported, not independently verified)"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Method: Proximal says (reported 9 Oct 2026, self-reported) it reached US$200M annualised revenue in under a year, for RL environments and post-training data for coding agents sold to frontier labs; the figure is a seed-stage claim not independently verified. Central US$200M, low US$120M (customer concentration and gross/net uncertainty), high US$250M. Split is an estimate (70% environments, 30% expert data). Full pass: Startup Fortune says the company reached this within roughly ten months of operating, names no customer and does not say whether any lab confirmed the figure. Main input: Startup Fortune (T3). Share = company ÷ function total; low and high move only this company's own range and hold the others at central. Proximal expert data for post-training: US$60m (18-112m) = run-rate US$200m (120-250m) × 30% (the smaller parts of a mixed vendor are varied by ±50% and the largest part absorbs the difference). Function total, annualised US$m (central): 7,084, made of identified candidates 6,934 and an allowance of 150 for vendors outside the register (for example Deccan AI, Shaip, Cogito Tech, Sigma AI, Centaur Labs, Revelo, e2f, Patronus AI, Haize Labs, Lightly, Mostly AI, Abundant, Ethos, Datumo and smaller environment builders). Largest, US$m: mercor 1,650, surge-ai 1,296, handshake-ai 882, micro1 850, scale-ai 412, other candidates 1,844, vendors outside the register 150. Full table: a local evidence capture. Share 0.847% (range 0.256-1.576%). Source published date: 2026-10-09 (full date)."},"labelling-multimodal":{"participates":false,"evidence":{"url":"https://startupfortune.com/proximal-reveals-200-million-revenue-run-rate-and-15-million-seed-round/","title":"Proximal reveals $200 million revenue run rate and $15 million seed round","publisher":"Startup Fortune","publishedDate":"2026-10-09","accessedDate":"2026-10-22","tier":"T3","locator":"Proximal reports $200 million in annualized revenue built in under a year; builds RL environments and post-training data for AI coding agents; customers are the labs building those agents; came out of stealth on September 29 (self-reported, not independently verified); no large-scale labelling and multimodal data offering described"}},"rl-environments":{"low":3.74,"central":7.613,"high":11.118,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"Annualised run-rate, Q3 2026 (company claim)","periodEnd":"2026-09-30","source":{"url":"https://startupfortune.com/proximal-reveals-200-million-revenue-run-rate-and-15-million-seed-round/","title":"Proximal reveals $200 million revenue run rate and $15 million seed round","publisher":"Startup Fortune","publishedDate":"2026-10-09","accessedDate":"2026-10-22","tier":"T5","locator":"Proximal reports $200 million in annualized revenue within roughly ten months of operating; came out of stealth on September 29; customers are the labs building coding agents, none named (self-reported, not independently verified)"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Method: Proximal says (reported 9 Oct 2026, self-reported) it reached US$200M annualised revenue in under a year, for RL environments and post-training data for coding agents sold to frontier labs; the figure is a seed-stage claim not independently verified. Central US$200M, low US$120M (customer concentration and gross/net uncertainty), high US$250M. Split is an estimate (70% environments, 30% expert data). Full pass: Startup Fortune says the company reached this within roughly ten months of operating, names no customer and does not say whether any lab confirmed the figure. Main input: Startup Fortune (T3). Share = company ÷ function total; low and high move only this company's own range and hold the others at central. Proximal RL environments and agent tasks: US$140m (66-212m) = run-rate US$200m (120-250m) × 70% (the smaller parts of a mixed vendor are varied by ±50% and the largest part absorbs the difference). Function total, annualised US$m (central): 1,839, made of identified candidates 1,739 and an allowance of 100 for vendors outside the register (for example Deccan AI, Shaip, Cogito Tech, Sigma AI, Centaur Labs, Revelo, e2f, Patronus AI, Haize Labs, Lightly, Mostly AI, Abundant, Ethos, Datumo and smaller environment builders). Largest, US$m: mercor 484, surge-ai 414, proximal 140, handshake-ai 136, micro1 130, other candidates 434, vendors outside the register 100. Full table: a local evidence capture. Share 7.613% (range 3.74-11.118%). Source published date: 2026-10-09 (full date)."},"evaluation":{"participates":false,"evidence":{"url":"https://startupfortune.com/proximal-reveals-200-million-revenue-run-rate-and-15-million-seed-round/","title":"Proximal reveals $200 million revenue run rate and $15 million seed round","publisher":"Startup Fortune","publishedDate":"2026-10-09","accessedDate":"2026-10-22","tier":"T3","locator":"Proximal reports $200 million in annualized revenue built in under a year; builds RL environments and post-training data for AI coding agents; customers are the labs building those agents; came out of stealth on September 29 (self-reported, not independently verified); no independent model evaluation offering described"}},"data-tooling":{"participates":false,"evidence":{"url":"https://startupfortune.com/proximal-reveals-200-million-revenue-run-rate-and-15-million-seed-round/","title":"Proximal reveals $200 million revenue run rate and $15 million seed round","publisher":"Startup Fortune","publishedDate":"2026-10-09","accessedDate":"2026-10-22","tier":"T3","locator":"Proximal reports $200 million in annualized revenue built in under a year; builds RL environments and post-training data for AI coding agents; customers are the labs building those agents; came out of stealth on September 29 (self-reported, not independently verified); no labelling and data software offering described"}}},"priorShares":{"expert-data":{"low":0,"central":0,"high":0,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"September 2025, before operations","periodEnd":"2025-09-30","source":{"url":"https://startupfortune.com/proximal-reveals-200-million-revenue-run-rate-and-15-million-seed-round/","title":"Proximal reveals $200 million revenue run rate and $15 million seed round","publisher":"Startup Fortune","publishedDate":"2026-10-09","accessedDate":"2026-10-22","tier":"T5","locator":"Reaches $200 million in annualized revenue within roughly ten months of operating; came out of stealth on September 29 (2026)"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Prior share in the Q3 2025 frame. Zero: the company says it reached US$200M within roughly ten months of operating (Startup Fortune, 9 Oct 2026), so it was not yet operating in September 2025 (inference from its stated age). Proximal expert data for post-training: US$0m (0-0m) = prior run-rate US$0m (0-0m) × 30% (split and the ±50% variation of the smaller parts held as today). Prior part total US$2,983m = the sum over all vendors of Q3 2025 run-rate × split, plus the allowance of 150 ÷ 1.5 = 100. Q3 2025 frame: each vendor's run-rate at the time, documented where it exists (Mercor US$500M, Handshake AI US$80M, micro1 US$45M, Snorkel US$20.8M, Arena and Proximal none, Appen, TaskUs and Innodata from their H1 or Q2 2025 reports scaled to today's basis, Surge US$1.2bn, Scale's data business US$850M, Turing US$337.5M, Fleet US$1M, Labelbox one fifth of today's, RWS today's divided by 1.45), and every other vendor at today's run-rate divided by 1.5 (the rounded median, 1.54, of the 12 documented multiples); each vendor's split across the five parts is held at today's. The frame is an estimate and its uncertainty is not in the share ranges. Share 0% (range 0-0%). The current share is 0.847% on the same basis, so the change is +0.8 points."},"rl-environments":{"low":0,"central":0,"high":0,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"September 2025, before operations","periodEnd":"2025-09-30","source":{"url":"https://startupfortune.com/proximal-reveals-200-million-revenue-run-rate-and-15-million-seed-round/","title":"Proximal reveals $200 million revenue run rate and $15 million seed round","publisher":"Startup Fortune","publishedDate":"2026-10-09","accessedDate":"2026-10-22","tier":"T5","locator":"Reaches $200 million in annualized revenue within roughly ten months of operating; came out of stealth on September 29 (2026)"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Prior share in the Q3 2025 frame. Zero: the company says it reached US$200M within roughly ten months of operating (Startup Fortune, 9 Oct 2026), so it was not yet operating in September 2025 (inference from its stated age). Proximal RL environments and agent tasks: US$0m (0-0m) = prior run-rate US$0m (0-0m) × 70% (split and the ±50% variation of the smaller parts held as today). Prior part total US$706m = the sum over all vendors of Q3 2025 run-rate × split, plus the allowance of 100 ÷ 1.5 = 66.7. Q3 2025 frame: each vendor's run-rate at the time, documented where it exists (Mercor US$500M, Handshake AI US$80M, micro1 US$45M, Snorkel US$20.8M, Arena and Proximal none, Appen, TaskUs and Innodata from their H1 or Q2 2025 reports scaled to today's basis, Surge US$1.2bn, Scale's data business US$850M, Turing US$337.5M, Fleet US$1M, Labelbox one fifth of today's, RWS today's divided by 1.45), and every other vendor at today's run-rate divided by 1.5 (the rounded median, 1.54, of the 12 documented multiples); each vendor's split across the five parts is held at today's. The frame is an estimate and its uncertainty is not in the share ranges. Share 0% (range 0-0%). The current share is 7.613% on the same basis, so the change is +7.6 points."}},"activityGrowth":{"current":{"low":120000000,"central":200000000,"high":250000000,"unit":"USD","basis":"Annualised revenue run-rate in US dollars (company claim; gross or net not stated)","period":"Run-rate announced 29 Sep 2026","periodEnd":"2026-09-30","source":{"url":"https://startupfortune.com/proximal-reveals-200-million-revenue-run-rate-and-15-million-seed-round/","title":"Proximal reveals $200 million revenue run rate and $15 million seed round","publisher":"Startup Fortune","publishedDate":"2026-10-09","accessedDate":"2026-10-22","tier":"T3","locator":"Proximal reports $200 million in annualized revenue within roughly ten months of operating; came out of stealth on September 29; customers are the labs building coding agents, none named (self-reported, not independently verified)"},"note":"Self-reported (Startup Fortune, 9 Oct 2026): US$200M annualised; the range keeps the screening range (US$120-250M)."},"prior":{"low":0,"central":0,"high":0,"unit":"USD","basis":"Annualised revenue run-rate in US dollars (company claim; gross or net not stated)","period":"September 2025, before operations","periodEnd":"2025-09-30","source":{"url":"https://startupfortune.com/proximal-reveals-200-million-revenue-run-rate-and-15-million-seed-round/","title":"Proximal reveals $200 million revenue run rate and $15 million seed round","publisher":"Startup Fortune","publishedDate":"2026-10-09","accessedDate":"2026-10-22","tier":"T3","locator":"Proximal reports $200 million in annualized revenue within roughly ten months of operating; came out of stealth on September 29; customers are the labs building coding agents, none named (self-reported, not independently verified)"},"note":"Inference from the stated age: the article says the company passed US$200M within roughly ten months of operating (and \"built in under a year\"), so it was not yet operating twelve months before the announcement. This is a zero for documented non-activity, not a missing figure, but it rests on a single self-reported article."}},"baselineRevenueUsd":{"low":200000000,"central":200000000,"high":200000000,"unit":"USD","basis":"Annualised revenue run-rate in US dollars (company claim)","period":"Run-rate announced 29 Sep 2026","periodEnd":"2026-09-30","source":{"url":"https://startupfortune.com/proximal-reveals-200-million-revenue-run-rate-and-15-million-seed-round/","title":"Proximal reveals $200 million revenue run rate and $15 million seed round","publisher":"Startup Fortune","publishedDate":"2026-10-09","accessedDate":"2026-10-22","tier":"T3","locator":"Proximal reports $200 million in annualized revenue within roughly ten months of operating; came out of stealth on September 29; customers are the labs building coding agents, none named (self-reported, not independently verified)"},"note":"Self-reported; not independently verified."},"screening":{"expert-data":{"levelLow":1,"levelHigh":2},"rl-environments":{"levelLow":1,"levelHigh":2}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"fd1dc13f12879d51b92006d2e1ea75a2a5fb37b4109c3c3ed0484e9248067bf6"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"fd1dc13f12879d51b92006d2e1ea75a2a5fb37b4109c3c3ed0484e9248067bf6"}}},"labelbox":{"schema":"5lc-index-inputs/1","key":"models-data-evaluation","entityId":"labelbox","name":"Labelbox","kind":"private","commercialActivityVerified":true,"shares":{"expert-data":{"low":0.342,"central":1.27,"high":4.375,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"Annualised run-rate, Q3 2026 (estimate)","periodEnd":"2026-09-30","source":{"url":"https://www.labelbox.com/","title":"Labelbox homepage","publisher":"Labelbox","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T5","locator":"Offerings: Horizon RL environments and evals, Alignerr expert network (2.6M+ experts), Terra robotics data; named customers include Meta and Google Cloud; Series E banner; no revenue"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Method: No revenue is published. The homepage lists Horizon RL environments and evals, the Alignerr expert network (2.6M+ experts), Terra robotics data and software, with Meta and Google Cloud as named customers and a US$100M+ Series E. Third-party databases give about US$50M for 2024 (not used: aggregators). Central US$150M, low US$60M, high US$400M. Split follows the card notes with 10% environments from Horizon. Full pass: the company says (Series E announcement of 14-15 Oct 2026, via Dealroom) that revenue grew more than fivefold year over year, it reached profitability in 2026 and it has 400+ customers including over 90% of leading US AI labs; no dollar figure, so the central stays at US$150M. Independent evaluation (added at the check-fix pass of 22 October 2026): the homepage lists Horizon 'RL environments and evals for post-training', says 'Over 90% of leading US AI labs post-train and evaluate on Labelbox', and names a Meta benchmark and a Google Cloud human-preference evaluation among its case studies. No split is published, so 5% of the run-rate is taken from expert data (65% to 60%) as a judgment. The Recursion eval tools on the same page serve companies' own AI applications and are outside this sublayer. Main input: Labelbox (T2). Share = company ÷ function total; low and high move only this company's own range and hold the others at central. Labelbox expert data for post-training: US$90m (24-320m) = run-rate US$150m (60-400m) × 60% (the smaller parts of a mixed vendor are varied by ±50% and the largest part absorbs the difference). Function total, annualised US$m (central): 7,084, made of identified candidates 6,934 and an allowance of 150 for vendors outside the register (for example Deccan AI, Shaip, Cogito Tech, Sigma AI, Centaur Labs, Revelo, e2f, Patronus AI, Haize Labs, Lightly, Mostly AI, Abundant, Ethos, Datumo and smaller environment builders). Largest, US$m: mercor 1,650, surge-ai 1,296, handshake-ai 882, micro1 850, scale-ai 412, other candidates 1,844, vendors outside the register 150. Full table: a local evidence capture. Share 1.27% (range 0.342-4.375%). Source published date: undated (undated)."},"labelling-multimodal":{"participates":false,"evidence":{"url":"https://www.labelbox.com/","title":"Labelbox homepage","publisher":"Labelbox","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T2","locator":"Offerings: Horizon RL environments and evals, Alignerr expert network (2.6M+ experts), Terra robotics data; named customers include Meta and Google Cloud; Series E banner; no revenue; no large-scale labelling and multimodal data offering described"}},"rl-environments":{"low":0.164,"central":0.816,"high":3.185,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"Annualised run-rate, Q3 2026 (estimate)","periodEnd":"2026-09-30","source":{"url":"https://www.labelbox.com/","title":"Labelbox homepage","publisher":"Labelbox","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T5","locator":"Offerings: Horizon RL environments and evals, Alignerr expert network (2.6M+ experts), Terra robotics data; named customers include Meta and Google Cloud; Series E banner; no revenue"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Method: No revenue is published. The homepage lists Horizon RL environments and evals, the Alignerr expert network (2.6M+ experts), Terra robotics data and software, with Meta and Google Cloud as named customers and a US$100M+ Series E. Third-party databases give about US$50M for 2024 (not used: aggregators). Central US$150M, low US$60M, high US$400M. Split follows the card notes with 10% environments from Horizon. Full pass: the company says (Series E announcement of 14-15 Oct 2026, via Dealroom) that revenue grew more than fivefold year over year, it reached profitability in 2026 and it has 400+ customers including over 90% of leading US AI labs; no dollar figure, so the central stays at US$150M. Independent evaluation (added at the check-fix pass of 22 October 2026): the homepage lists Horizon 'RL environments and evals for post-training', says 'Over 90% of leading US AI labs post-train and evaluate on Labelbox', and names a Meta benchmark and a Google Cloud human-preference evaluation among its case studies. No split is published, so 5% of the run-rate is taken from expert data (65% to 60%) as a judgment. The Recursion eval tools on the same page serve companies' own AI applications and are outside this sublayer. Main input: Labelbox (T2). Share = company ÷ function total; low and high move only this company's own range and hold the others at central. Labelbox RL environments and agent tasks: US$15m (3-60m) = run-rate US$150m (60-400m) × 10% (the smaller parts of a mixed vendor are varied by ±50% and the largest part absorbs the difference). Function total, annualised US$m (central): 1,839, made of identified candidates 1,739 and an allowance of 100 for vendors outside the register (for example Deccan AI, Shaip, Cogito Tech, Sigma AI, Centaur Labs, Revelo, e2f, Patronus AI, Haize Labs, Lightly, Mostly AI, Abundant, Ethos, Datumo and smaller environment builders). Largest, US$m: mercor 484, surge-ai 414, proximal 140, handshake-ai 136, micro1 130, other candidates 434, vendors outside the register 100. Full table: a local evidence capture. Share 0.816% (range 0.164-3.185%). Source published date: undated (undated)."},"evaluation":{"low":0.259,"central":1.28,"high":4.931,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"Annualised run-rate, Q3 2026 (estimate)","periodEnd":"2026-09-30","source":{"url":"https://www.labelbox.com/","title":"Labelbox homepage","publisher":"Labelbox","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T5","locator":"Horizon: 'RL environments and evals for post-training, across reasoning, tool use and computer use'; 'Over 90% of leading US AI labs post-train and evaluate on Labelbox'; case studies: Meta (benchmark) and Google Cloud (human preference signal for evaluating LLMs); no revenue"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Method: No revenue is published. The homepage lists Horizon RL environments and evals, the Alignerr expert network (2.6M+ experts), Terra robotics data and software, with Meta and Google Cloud as named customers and a US$100M+ Series E. Third-party databases give about US$50M for 2024 (not used: aggregators). Central US$150M, low US$60M, high US$400M. Split follows the card notes with 10% environments from Horizon. Full pass: the company says (Series E announcement of 14-15 Oct 2026, via Dealroom) that revenue grew more than fivefold year over year, it reached profitability in 2026 and it has 400+ customers including over 90% of leading US AI labs; no dollar figure, so the central stays at US$150M. Independent evaluation (added at the check-fix pass of 22 October 2026): the homepage lists Horizon 'RL environments and evals for post-training', says 'Over 90% of leading US AI labs post-train and evaluate on Labelbox', and names a Meta benchmark and a Google Cloud human-preference evaluation among its case studies. No split is published, so 5% of the run-rate is taken from expert data (65% to 60%) as a judgment. The Recursion eval tools on the same page serve companies' own AI applications and are outside this sublayer. Main input: Labelbox (T2). Share = company ÷ function total; low and high move only this company's own range and hold the others at central. Labelbox independent model evaluation: US$7.5m (1.5-30m) = run-rate US$150m (60-400m) × 5% (the smaller parts of a mixed vendor are varied by ±50% and the largest part absorbs the difference). Function total, annualised US$m (central): 586, made of identified candidates 566 and an allowance of 20 for vendors outside the register (for example Deccan AI, Shaip, Cogito Tech, Sigma AI, Centaur Labs, Revelo, e2f, Patronus AI, Haize Labs, Lightly, Mostly AI, Abundant, Ethos, Datumo and smaller environment builders). Largest, US$m: arena-ai 110, surge-ai 90, mercor 66, scale-ai 45, innodata 33, other candidates 222, vendors outside the register 20. Full table: a local evidence capture. Share 1.28% (range 0.259-4.931%). Source published date: undated (undated)."},"data-tooling":{"low":3.21,"central":14.223,"high":39.878,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"Annualised run-rate, Q3 2026 (estimate)","periodEnd":"2026-09-30","source":{"url":"https://www.labelbox.com/","title":"Labelbox homepage","publisher":"Labelbox","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T5","locator":"Offerings: Horizon RL environments and evals, Alignerr expert network (2.6M+ experts), Terra robotics data; named customers include Meta and Google Cloud; Series E banner; no revenue"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Method: No revenue is published. The homepage lists Horizon RL environments and evals, the Alignerr expert network (2.6M+ experts), Terra robotics data and software, with Meta and Google Cloud as named customers and a US$100M+ Series E. Third-party databases give about US$50M for 2024 (not used: aggregators). Central US$150M, low US$60M, high US$400M. Split follows the card notes with 10% environments from Horizon. Full pass: the company says (Series E announcement of 14-15 Oct 2026, via Dealroom) that revenue grew more than fivefold year over year, it reached profitability in 2026 and it has 400+ customers including over 90% of leading US AI labs; no dollar figure, so the central stays at US$150M. Independent evaluation (added at the check-fix pass of 22 October 2026): the homepage lists Horizon 'RL environments and evals for post-training', says 'Over 90% of leading US AI labs post-train and evaluate on Labelbox', and names a Meta benchmark and a Google Cloud human-preference evaluation among its case studies. No split is published, so 5% of the run-rate is taken from expert data (65% to 60%) as a judgment. The Recursion eval tools on the same page serve companies' own AI applications and are outside this sublayer. Main input: Labelbox (T2). Share = company ÷ function total; low and high move only this company's own range and hold the others at central. Labelbox labelling and data software: US$37.5m (7.5-150m) = run-rate US$150m (60-400m) × 25% (the smaller parts of a mixed vendor are varied by ±50% and the largest part absorbs the difference). Function total, annualised US$m (central): 264, made of identified candidates 234 and an allowance of 30 for vendors outside the register (for example Deccan AI, Shaip, Cogito Tech, Sigma AI, Centaur Labs, Revelo, e2f, Patronus AI, Haize Labs, Lightly, Mostly AI, Abundant, Ethos, Datumo and smaller environment builders). Largest, US$m: snorkel-ai 37.5, labelbox 37.5, encord 36, roboflow 25, superannotate 21, other candidates 76.7, vendors outside the register 30. Full table: a local evidence capture. Share 14.223% (range 3.21-39.878%). Source published date: undated (undated)."}},"activityGrowth":{"current":{"low":5,"central":5,"high":5,"unit":"index","basis":"Revenue index, revenue a year earlier = 1 (the company says growth was more than fivefold year over year; the exact multiple is not published)","period":"Year to October 2026","periodEnd":"2026-10-14","source":{"url":"https://dealroom.co/news/162716-labelbox-lands-100m-series-e-to-push-reinforcement-learning-and-ai-agent","title":"Labelbox lands $100M Series E to push reinforcement learning and AI agents","publisher":"Dealroom (reporting a Labelbox announcement)","publishedDate":"2026-10-14","accessedDate":"2026-10-22","tier":"T2","locator":"The company says it grew more than fivefold year over year and reached profitability in 2026; works with more than 400 customers, including over 90% of leading US AI labs; no dollar revenue"},"note":"The company says revenue grew more than fivefold year over year (Series E announcement, 14 Oct 2026, via Dealroom); 5 is the stated lower bound. Growth is at least 5 times, above the engine's +300% cap."},"prior":{"low":1,"central":1,"high":1,"unit":"index","basis":"Revenue index, revenue a year earlier = 1 (the company says growth was more than fivefold year over year; the exact multiple is not published)","period":"Year to October 2025","periodEnd":"2025-10-14","source":{"url":"https://dealroom.co/news/162716-labelbox-lands-100m-series-e-to-push-reinforcement-learning-and-ai-agent","title":"Labelbox lands $100M Series E to push reinforcement learning and AI agents","publisher":"Dealroom (reporting a Labelbox announcement)","publishedDate":"2026-10-14","accessedDate":"2026-10-22","tier":"T2","locator":"The company says it grew more than fivefold year over year and reached profitability in 2026; works with more than 400 customers, including over 90% of leading US AI labs; no dollar revenue"},"note":"Base of the index (revenue a year earlier = 1)."}},"screening":{"expert-data":{"levelLow":1,"levelHigh":3},"rl-environments":{"levelLow":1,"levelHigh":2},"data-tooling":{"levelLow":1,"levelHigh":2},"evaluation":{"levelLow":1,"levelHigh":2}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"e3c608f974d3503e2f2bfe2c6d5e90eed79abc1cb02a4ccc8b88d56a48cd806b"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"e3c608f974d3503e2f2bfe2c6d5e90eed79abc1cb02a4ccc8b88d56a48cd806b"}}},"transperfect-dataforce":{"schema":"5lc-index-inputs/1","key":"models-data-evaluation","entityId":"transperfect-dataforce","name":"TransPerfect DataForce","kind":"unit","parentId":"transperfect","commercialActivityVerified":false,"shares":{"expert-data":{"low":0.071,"central":0.339,"high":1.259,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"Mid-2026 (estimate)","periodEnd":"2026-09-30","source":{"url":"https://www.dataforce.ai/","title":"DataForce homepage","publisher":"DataForce (TransPerfect)","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T5","locator":"Generative AI training, data collection and annotation; 1M+ remote contributors; logo wall includes Amazon, ByteDance, Microsoft; no revenue"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Method: DataForce is TransPerfect's AI-data division (1M+ contributors); TransPerfect's own release reports billed revenue of US$1.23bn for 2024 (headline seen in search results; the page text was not available) and DataForce's share is not published. Central US$120M is an estimate (US$50-300M). Unit only. Main input: DataForce (TransPerfect) (T5). Share = company ÷ function total; low and high move only this company's own range and hold the others at central. TransPerfect DataForce expert data for post-training: US$24m (5-90m) = run-rate US$120m (50-300m) × 20% (the smaller parts of a mixed vendor are varied by ±50% and the largest part absorbs the difference). Function total, annualised US$m (central): 7,084, made of identified candidates 6,934 and an allowance of 150 for vendors outside the register (for example Deccan AI, Shaip, Cogito Tech, Sigma AI, Centaur Labs, Revelo, e2f, Patronus AI, Haize Labs, Lightly, Mostly AI, Abundant, Ethos, Datumo and smaller environment builders). Largest, US$m: mercor 1,650, surge-ai 1,296, handshake-ai 882, micro1 850, scale-ai 412, other candidates 1,844, vendors outside the register 150. Full table: a local evidence capture. Share 0.339% (range 0.071-1.259%). Source published date: undated (undated)."},"labelling-multimodal":{"low":2.061,"central":5.457,"high":13.966,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"Mid-2026 (estimate)","periodEnd":"2026-09-30","source":{"url":"https://www.dataforce.ai/","title":"DataForce homepage","publisher":"DataForce (TransPerfect)","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T5","locator":"Generative AI training, data collection and annotation; 1M+ remote contributors; logo wall includes Amazon, ByteDance, Microsoft; no revenue"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Method: DataForce is TransPerfect's AI-data division (1M+ contributors); TransPerfect's own release reports billed revenue of US$1.23bn for 2024 (headline seen in search results; the page text was not available) and DataForce's share is not published. Central US$120M is an estimate (US$50-300M). Unit only. Main input: DataForce (TransPerfect) (T5). Share = company ÷ function total; low and high move only this company's own range and hold the others at central. TransPerfect DataForce large-scale labelling and multimodal data: US$96m (35-270m) = run-rate US$120m (50-300m) × 80% (the smaller parts of a mixed vendor are varied by ±50% and the largest part absorbs the difference). Function total, annualised US$m (central): 1,759, made of identified candidates 1,609 and an allowance of 150 for vendors outside the register (for example Deccan AI, Shaip, Cogito Tech, Sigma AI, Centaur Labs, Revelo, e2f, Patronus AI, Haize Labs, Lightly, Mostly AI, Abundant, Ethos, Datumo and smaller environment builders). Largest, US$m: scale-ai 262, appen 176, taskus 146, rws-trainai 140, telus-digital 124, other candidates 762, vendors outside the register 150. Full table: a local evidence capture. Share 5.457% (range 2.061-13.966%). Source published date: undated (undated)."},"rl-environments":{"participates":false,"evidence":{"url":"https://www.dataforce.ai/","title":"DataForce homepage","publisher":"DataForce (TransPerfect)","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T2","locator":"Generative AI training, data collection and annotation; 1M+ remote contributors; logo wall includes Amazon, ByteDance, Microsoft; no revenue; no RL environments and agent tasks offering described"}},"evaluation":{"participates":false,"evidence":{"url":"https://www.dataforce.ai/","title":"DataForce homepage","publisher":"DataForce (TransPerfect)","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T2","locator":"Generative AI training, data collection and annotation; 1M+ remote contributors; logo wall includes Amazon, ByteDance, Microsoft; no revenue; no independent model evaluation offering described"}},"data-tooling":{"participates":false,"evidence":{"url":"https://www.dataforce.ai/","title":"DataForce homepage","publisher":"DataForce (TransPerfect)","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T2","locator":"Generative AI training, data collection and annotation; 1M+ remote contributors; logo wall includes Amazon, ByteDance, Microsoft; no revenue; no labelling and data software offering described"}}},"screening":{"expert-data":{"levelLow":1,"levelHigh":2},"labelling-multimodal":{"levelLow":1,"levelHigh":2}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"5ba4c9b8005f9f43be04ef9243417161f1b204107118e35a5b61068e6b30c9b2"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"5ba4c9b8005f9f43be04ef9243417161f1b204107118e35a5b61068e6b30c9b2"}}},"toloka":{"schema":"5lc-index-inputs/1","key":"models-data-evaluation","entityId":"toloka","name":"Toloka","kind":"private","commercialActivityVerified":true,"shares":{"expert-data":{"low":0.313,"central":1.087,"high":2.943,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"Q1 2025 evidence (growth over 100%), extrapolated to a 2026 run-rate (estimate)","periodEnd":"2025-03-31","source":{"url":"https://www.sec.gov/Archives/edgar/data/1513845/000110465925050975/tm2515580d1_ex99-2.htm","title":"Nebius Group Q1 2025 letter to shareholders","publisher":"Nebius Group","publishedDate":"2025-05-20","accessedDate":"2026-10-22","tier":"T5","locator":"The company says Toloka's revenue more than doubled year over year in Q1 2025 and all major frontier lab contracts were renewed; customers include Anthropic, Amazon and Microsoft; no dollar figure"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Method: Nebius's shareholder letter (20 May 2025) says Toloka's revenue more than doubled year over year in Q1 2025 and lists Anthropic, Amazon and Microsoft as customers; no dollar figure is given. Third-party databases give US$19-50M (not used: aggregators). Central US$110M extrapolates, low US$40M, high US$250M. Mindrift is Toloka's own platform and is counted here. Evidence is over 12 months old, so the engine adds the staleness widening. Full pass: searches found no Toloka revenue after Q1 2025; Nebius's Q2 2026 results release (6-K) says Nebius no longer consolidates Toloka (equity method since a third-party investment completed in Q2 2025) and gives no Toloka figure. Main input: Nebius Group (T2). Share = company ÷ function total; low and high move only this company's own range and hold the others at central. Toloka expert data for post-training: US$77m (22-212m) = run-rate US$110m (40-250m) × 70% (the smaller parts of a mixed vendor are varied by ±50% and the largest part absorbs the difference). Function total, annualised US$m (central): 7,084, made of identified candidates 6,934 and an allowance of 150 for vendors outside the register (for example Deccan AI, Shaip, Cogito Tech, Sigma AI, Centaur Labs, Revelo, e2f, Patronus AI, Haize Labs, Lightly, Mostly AI, Abundant, Ethos, Datumo and smaller environment builders). Largest, US$m: mercor 1,650, surge-ai 1,296, handshake-ai 882, micro1 850, scale-ai 412, other candidates 1,844, vendors outside the register 150. Full table: a local evidence capture. Share 1.087% (range 0.313-2.943%). Source published date: 2025-05-20 (full date)."},"labelling-multimodal":{"low":0.23,"central":1.251,"high":4.138,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"Q1 2025 evidence (growth over 100%), extrapolated to a 2026 run-rate (estimate)","periodEnd":"2025-03-31","source":{"url":"https://www.sec.gov/Archives/edgar/data/1513845/000110465925050975/tm2515580d1_ex99-2.htm","title":"Nebius Group Q1 2025 letter to shareholders","publisher":"Nebius Group","publishedDate":"2025-05-20","accessedDate":"2026-10-22","tier":"T5","locator":"The company says Toloka's revenue more than doubled year over year in Q1 2025 and all major frontier lab contracts were renewed; customers include Anthropic, Amazon and Microsoft; no dollar figure"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Method: Nebius's shareholder letter (20 May 2025) says Toloka's revenue more than doubled year over year in Q1 2025 and lists Anthropic, Amazon and Microsoft as customers; no dollar figure is given. Third-party databases give US$19-50M (not used: aggregators). Central US$110M extrapolates, low US$40M, high US$250M. Mindrift is Toloka's own platform and is counted here. Evidence is over 12 months old, so the engine adds the staleness widening. Full pass: searches found no Toloka revenue after Q1 2025; Nebius's Q2 2026 results release (6-K) says Nebius no longer consolidates Toloka (equity method since a third-party investment completed in Q2 2025) and gives no Toloka figure. Main input: Nebius Group (T2). Share = company ÷ function total; low and high move only this company's own range and hold the others at central. Toloka large-scale labelling and multimodal data: US$22m (4-75m) = run-rate US$110m (40-250m) × 20% (the smaller parts of a mixed vendor are varied by ±50% and the largest part absorbs the difference). Function total, annualised US$m (central): 1,759, made of identified candidates 1,609 and an allowance of 150 for vendors outside the register (for example Deccan AI, Shaip, Cogito Tech, Sigma AI, Centaur Labs, Revelo, e2f, Patronus AI, Haize Labs, Lightly, Mostly AI, Abundant, Ethos, Datumo and smaller environment builders). Largest, US$m: scale-ai 262, appen 176, taskus 146, rws-trainai 140, telus-digital 124, other candidates 762, vendors outside the register 150. Full table: a local evidence capture. Share 1.251% (range 0.23-4.138%). Source published date: 2025-05-20 (full date)."},"rl-environments":{"participates":false,"evidence":{"url":"https://www.sec.gov/Archives/edgar/data/1513845/000110465925050975/tm2515580d1_ex99-2.htm","title":"Nebius Group Q1 2025 letter to shareholders","publisher":"Nebius Group","publishedDate":"2025-05-20","accessedDate":"2026-10-22","tier":"T2","locator":"The company says Toloka's revenue more than doubled year over year in Q1 2025 and all major frontier lab contracts were renewed; customers include Anthropic, Amazon and Microsoft; no dollar figure; no RL environments and agent tasks offering described"}},"evaluation":{"low":0.347,"central":1.878,"high":6.124,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"Q1 2025 evidence (growth over 100%), extrapolated to a 2026 run-rate (estimate)","periodEnd":"2025-03-31","source":{"url":"https://www.sec.gov/Archives/edgar/data/1513845/000110465925050975/tm2515580d1_ex99-2.htm","title":"Nebius Group Q1 2025 letter to shareholders","publisher":"Nebius Group","publishedDate":"2025-05-20","accessedDate":"2026-10-22","tier":"T5","locator":"The company says Toloka's revenue more than doubled year over year in Q1 2025 and all major frontier lab contracts were renewed; customers include Anthropic, Amazon and Microsoft; no dollar figure"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Method: Nebius's shareholder letter (20 May 2025) says Toloka's revenue more than doubled year over year in Q1 2025 and lists Anthropic, Amazon and Microsoft as customers; no dollar figure is given. Third-party databases give US$19-50M (not used: aggregators). Central US$110M extrapolates, low US$40M, high US$250M. Mindrift is Toloka's own platform and is counted here. Evidence is over 12 months old, so the engine adds the staleness widening. Full pass: searches found no Toloka revenue after Q1 2025; Nebius's Q2 2026 results release (6-K) says Nebius no longer consolidates Toloka (equity method since a third-party investment completed in Q2 2025) and gives no Toloka figure. Main input: Nebius Group (T2). Share = company ÷ function total; low and high move only this company's own range and hold the others at central. Toloka independent model evaluation: US$11m (2-37.5m) = run-rate US$110m (40-250m) × 10% (the smaller parts of a mixed vendor are varied by ±50% and the largest part absorbs the difference). Function total, annualised US$m (central): 586, made of identified candidates 566 and an allowance of 20 for vendors outside the register (for example Deccan AI, Shaip, Cogito Tech, Sigma AI, Centaur Labs, Revelo, e2f, Patronus AI, Haize Labs, Lightly, Mostly AI, Abundant, Ethos, Datumo and smaller environment builders). Largest, US$m: arena-ai 110, surge-ai 90, mercor 66, scale-ai 45, innodata 33, other candidates 222, vendors outside the register 20. Full table: a local evidence capture. Share 1.878% (range 0.347-6.124%). Source published date: 2025-05-20 (full date)."},"data-tooling":{"participates":false,"evidence":{"url":"https://www.sec.gov/Archives/edgar/data/1513845/000110465925050975/tm2515580d1_ex99-2.htm","title":"Nebius Group Q1 2025 letter to shareholders","publisher":"Nebius Group","publishedDate":"2025-05-20","accessedDate":"2026-10-22","tier":"T2","locator":"The company says Toloka's revenue more than doubled year over year in Q1 2025 and all major frontier lab contracts were renewed; customers include Anthropic, Amazon and Microsoft; no dollar figure; no labelling and data software offering described"}}},"screening":{"expert-data":{"levelLow":1,"levelHigh":3},"labelling-multimodal":{"levelLow":1,"levelHigh":2},"evaluation":{"levelLow":1,"levelHigh":2}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"7da8bb9fd4155878562b3c369acad988f3b0ce6df681ae06d519bba203895307"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"7da8bb9fd4155878562b3c369acad988f3b0ce6df681ae06d519bba203895307"}}},"fleet-ai":{"schema":"5lc-index-inputs/1","key":"models-data-evaluation","entityId":"fleet-ai","name":"Fleet","kind":"private","commercialActivityVerified":true,"shares":{"expert-data":{"participates":false,"evidence":{"url":"https://sacra.com/c/fleet","title":"Fleet revenue, funding and news","publisher":"Sacra","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T4","locator":"Sacra estimates Fleet reached $60M in annualized revenue in April 2026 (latest quarter x 4), up from $1M in 2025, driven by frontier lab demand for RL environments; no expert data for post-training offering described"}},"labelling-multimodal":{"participates":false,"evidence":{"url":"https://sacra.com/c/fleet","title":"Fleet revenue, funding and news","publisher":"Sacra","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T4","locator":"Sacra estimates Fleet reached $60M in annualized revenue in April 2026 (latest quarter x 4), up from $1M in 2025, driven by frontier lab demand for RL environments; no large-scale labelling and multimodal data offering described"}},"rl-environments":{"low":3.335,"central":5.438,"high":10.315,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"Annualised run-rate, Q3 2026 (estimate)","periodEnd":"2026-09-30","source":{"url":"https://sacra.com/c/fleet","title":"Fleet revenue, funding and news","publisher":"Sacra","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T5","locator":"Sacra estimates Fleet reached $60M in annualized revenue in April 2026 (latest quarter x 4), up from $1M in 2025"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Method: Sacra (T4) estimates US$60M annualised in April 2026; RuntimeWire (T4, unnamed sources, 1 Jun 2026) reports US$63M and US$160M tracked for the next quarter. Central US$100M, low US$60M (the April estimate), high US$200M. The company's own page describes its work only in general terms. Full pass: no new figure beyond Sacra and RuntimeWire; neither gives a date for the earlier US$1M. Main input: Sacra (T4). Share = company ÷ function total; low and high move only this company's own range and hold the others at central. Fleet RL environments and agent tasks: US$100m (60-200m) = run-rate US$100m (60-200m) × 100%. Function total, annualised US$m (central): 1,839, made of identified candidates 1,739 and an allowance of 100 for vendors outside the register (for example Deccan AI, Shaip, Cogito Tech, Sigma AI, Centaur Labs, Revelo, e2f, Patronus AI, Haize Labs, Lightly, Mostly AI, Abundant, Ethos, Datumo and smaller environment builders). Largest, US$m: mercor 484, surge-ai 414, proximal 140, handshake-ai 136, micro1 130, other candidates 434, vendors outside the register 100. Full table: a local evidence capture. Share 5.438% (range 3.335-10.315%). Second source(s): RuntimeWire 2026-06-01 (T4): Reports run-rate climbing from $1M six months earlier to $63M and tracking $160M next quarter, from unnamed sources; $45M Series A at $725M. Source published date: undated (undated)."},"evaluation":{"participates":false,"evidence":{"url":"https://sacra.com/c/fleet","title":"Fleet revenue, funding and news","publisher":"Sacra","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T4","locator":"Sacra estimates Fleet reached $60M in annualized revenue in April 2026 (latest quarter x 4), up from $1M in 2025, driven by frontier lab demand for RL environments; no independent model evaluation offering described"}},"data-tooling":{"participates":false,"evidence":{"url":"https://sacra.com/c/fleet","title":"Fleet revenue, funding and news","publisher":"Sacra","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T4","locator":"Sacra estimates Fleet reached $60M in annualized revenue in April 2026 (latest quarter x 4), up from $1M in 2025, driven by frontier lab demand for RL environments; no labelling and data software offering described"}}},"baselineRevenueUsd":{"low":60000000,"central":60000000,"high":60000000,"unit":"USD","basis":"Annualised revenue run-rate in US dollars (latest quarter × 4)","period":"April 2026 run-rate","periodEnd":"2026-04-30","source":{"url":"https://sacra.com/c/fleet","title":"Fleet revenue, funding and news","publisher":"Sacra","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T4","locator":"Sacra estimates Fleet reached $60M in annualized revenue in April 2026 (latest quarter x 4), up from $1M in 2025"},"note":"Sacra estimate (T4); RuntimeWire reported US$63M on 1 Jun 2026 (T4)."},"screening":{"rl-environments":{"levelLow":1,"levelHigh":2}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"1a4ee5f526e16943ed81d27f42fc99ec99231dc0fee817c0a02da628f48a91f1"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"1a4ee5f526e16943ed81d27f42fc99ec99231dc0fee817c0a02da628f48a91f1"}}},"uber-ai-solutions":{"schema":"5lc-index-inputs/1","key":"models-data-evaluation","entityId":"uber-ai-solutions","name":"Uber AI Solutions","kind":"unit","parentId":"uber","commercialActivityVerified":true,"shares":{"expert-data":{"low":0.042,"central":0.339,"high":1.568,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"Mid-2026 (estimate)","periodEnd":"2026-06-30","source":{"url":"https://startupfortune.com/uber-fires-the-two-executives-who-built-its-ai-data-labeling-bet/","title":"Uber fires the two executives who built its AI data labeling bet","publisher":"Startup Fortune","publishedDate":"2026-07-01","accessedDate":"2026-10-22","tier":"T5","locator":"Uber has not disclosed revenue for AI Solutions; customers include Alphabet, Aurora Innovation and Niantic (per Forbes); service expanded to more than 30 countries"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Method: Uber has not disclosed revenue for AI Solutions. The service operates in more than 30 countries and counts Alphabet, Aurora Innovation and Niantic among its customers (Forbes, via Startup Fortune, 1 Jul 2026); two senior leaders left in June 2026. Central US$80M is a 5LayerCake estimate with a wide range (US$20-250M). Unit only, never Uber's total. Full pass: searches found no revenue figure for the unit. Main input: Startup Fortune (T3). Share = company ÷ function total; low and high move only this company's own range and hold the others at central. Uber AI Solutions expert data for post-training: US$24m (3-112m) = run-rate US$80m (20-250m) × 30% (the smaller parts of a mixed vendor are varied by ±50% and the largest part absorbs the difference). Function total, annualised US$m (central): 7,084, made of identified candidates 6,934 and an allowance of 150 for vendors outside the register (for example Deccan AI, Shaip, Cogito Tech, Sigma AI, Centaur Labs, Revelo, e2f, Patronus AI, Haize Labs, Lightly, Mostly AI, Abundant, Ethos, Datumo and smaller environment builders). Largest, US$m: mercor 1,650, surge-ai 1,296, handshake-ai 882, micro1 850, scale-ai 412, other candidates 1,844, vendors outside the register 150. Full table: a local evidence capture. Share 0.339% (range 0.042-1.568%). Source published date: 2026-07-01 (full date)."},"labelling-multimodal":{"low":0.642,"central":3.183,"high":11.092,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"Mid-2026 (estimate)","periodEnd":"2026-06-30","source":{"url":"https://startupfortune.com/uber-fires-the-two-executives-who-built-its-ai-data-labeling-bet/","title":"Uber fires the two executives who built its AI data labeling bet","publisher":"Startup Fortune","publishedDate":"2026-07-01","accessedDate":"2026-10-22","tier":"T5","locator":"Uber has not disclosed revenue for AI Solutions; customers include Alphabet, Aurora Innovation and Niantic (per Forbes); service expanded to more than 30 countries"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Method: Uber has not disclosed revenue for AI Solutions. The service operates in more than 30 countries and counts Alphabet, Aurora Innovation and Niantic among its customers (Forbes, via Startup Fortune, 1 Jul 2026); two senior leaders left in June 2026. Central US$80M is a 5LayerCake estimate with a wide range (US$20-250M). Unit only, never Uber's total. Full pass: searches found no revenue figure for the unit. Main input: Startup Fortune (T3). Share = company ÷ function total; low and high move only this company's own range and hold the others at central. Uber AI Solutions large-scale labelling and multimodal data: US$56m (11-212m) = run-rate US$80m (20-250m) × 70% (the smaller parts of a mixed vendor are varied by ±50% and the largest part absorbs the difference). Function total, annualised US$m (central): 1,759, made of identified candidates 1,609 and an allowance of 150 for vendors outside the register (for example Deccan AI, Shaip, Cogito Tech, Sigma AI, Centaur Labs, Revelo, e2f, Patronus AI, Haize Labs, Lightly, Mostly AI, Abundant, Ethos, Datumo and smaller environment builders). Largest, US$m: scale-ai 262, appen 176, taskus 146, rws-trainai 140, telus-digital 124, other candidates 762, vendors outside the register 150. Full table: a local evidence capture. Share 3.183% (range 0.642-11.092%). Source published date: 2026-07-01 (full date)."},"rl-environments":{"participates":false,"evidence":{"url":"https://startupfortune.com/uber-fires-the-two-executives-who-built-its-ai-data-labeling-bet/","title":"Uber fires the two executives who built its AI data labeling bet","publisher":"Startup Fortune","publishedDate":"2026-07-01","accessedDate":"2026-10-22","tier":"T3","locator":"Uber has not disclosed revenue for AI Solutions; customers include Alphabet, Aurora Innovation and Niantic (per Forbes); service expanded to more than 30 countries; no RL environments and agent tasks offering described"}},"evaluation":{"participates":false,"evidence":{"url":"https://startupfortune.com/uber-fires-the-two-executives-who-built-its-ai-data-labeling-bet/","title":"Uber fires the two executives who built its AI data labeling bet","publisher":"Startup Fortune","publishedDate":"2026-07-01","accessedDate":"2026-10-22","tier":"T3","locator":"Uber has not disclosed revenue for AI Solutions; customers include Alphabet, Aurora Innovation and Niantic (per Forbes); service expanded to more than 30 countries; no independent model evaluation offering described"}},"data-tooling":{"participates":false,"evidence":{"url":"https://startupfortune.com/uber-fires-the-two-executives-who-built-its-ai-data-labeling-bet/","title":"Uber fires the two executives who built its AI data labeling bet","publisher":"Startup Fortune","publishedDate":"2026-07-01","accessedDate":"2026-10-22","tier":"T3","locator":"Uber has not disclosed revenue for AI Solutions; customers include Alphabet, Aurora Innovation and Niantic (per Forbes); service expanded to more than 30 countries; no labelling and data software offering described"}}},"screening":{"expert-data":{"levelLow":1,"levelHigh":2},"labelling-multimodal":{"levelLow":1,"levelHigh":2}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"577823a87efac469800647204ae60b7a84fd13055927a8c140cc9859c85f0a42"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"577823a87efac469800647204ae60b7a84fd13055927a8c140cc9859c85f0a42"}}},"welo-data":{"schema":"5lc-index-inputs/1","key":"models-data-evaluation","entityId":"welo-data","name":"Welo Data","kind":"unit","parentId":"welocalize","commercialActivityVerified":false,"shares":{"expert-data":{"low":0.064,"central":0.339,"high":1.383,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"Mid-2026 (estimate)","periodEnd":"2026-09-30","source":{"url":"https://welodata.ai/","title":"Welo Data homepage","publisher":"Welo Data (Welocalize)","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T5","locator":"Offerings: data collection, annotation, multilingual RL environments, model evaluation and red teaming; 500k+ curated experts; case studies up to 2026-05-14; no revenue"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Method: Welo Data is Welocalize's AI-data brand (500k+ experts, multilingual RL environments, evaluation); no revenue is published and group figures are not used. Central US$80M is an estimate (US$30-220M). Unit only. Full pass: searches found no unit revenue figure. Main input: Welo Data (Welocalize) (T2). Share = company ÷ function total; low and high move only this company's own range and hold the others at central. Welo Data expert data for post-training: US$24m (4.5-99m) = run-rate US$80m (30-220m) × 30% (the smaller parts of a mixed vendor are varied by ±50% and the largest part absorbs the difference). Function total, annualised US$m (central): 7,084, made of identified candidates 6,934 and an allowance of 150 for vendors outside the register (for example Deccan AI, Shaip, Cogito Tech, Sigma AI, Centaur Labs, Revelo, e2f, Patronus AI, Haize Labs, Lightly, Mostly AI, Abundant, Ethos, Datumo and smaller environment builders). Largest, US$m: mercor 1,650, surge-ai 1,296, handshake-ai 882, micro1 850, scale-ai 412, other candidates 1,844, vendors outside the register 150. Full table: a local evidence capture. Share 0.339% (range 0.064-1.383%). Source published date: undated (undated)."},"labelling-multimodal":{"low":0.565,"central":2.501,"high":9.041,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"Mid-2026 (estimate)","periodEnd":"2026-09-30","source":{"url":"https://welodata.ai/","title":"Welo Data homepage","publisher":"Welo Data (Welocalize)","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T5","locator":"Offerings: data collection, annotation, multilingual RL environments, model evaluation and red teaming; 500k+ curated experts; case studies up to 2026-05-14; no revenue"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Method: Welo Data is Welocalize's AI-data brand (500k+ experts, multilingual RL environments, evaluation); no revenue is published and group figures are not used. Central US$80M is an estimate (US$30-220M). Unit only. Full pass: searches found no unit revenue figure. Main input: Welo Data (Welocalize) (T2). Share = company ÷ function total; low and high move only this company's own range and hold the others at central. Welo Data large-scale labelling and multimodal data: US$44m (9.75-170m) = run-rate US$80m (30-220m) × 55% (the smaller parts of a mixed vendor are varied by ±50% and the largest part absorbs the difference). Function total, annualised US$m (central): 1,759, made of identified candidates 1,609 and an allowance of 150 for vendors outside the register (for example Deccan AI, Shaip, Cogito Tech, Sigma AI, Centaur Labs, Revelo, e2f, Patronus AI, Haize Labs, Lightly, Mostly AI, Abundant, Ethos, Datumo and smaller environment builders). Largest, US$m: scale-ai 262, appen 176, taskus 146, rws-trainai 140, telus-digital 124, other candidates 762, vendors outside the register 150. Full table: a local evidence capture. Share 2.501% (range 0.565-9.041%). Source published date: undated (undated)."},"rl-environments":{"low":0.082,"central":0.435,"high":1.771,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"Mid-2026 (estimate)","periodEnd":"2026-09-30","source":{"url":"https://welodata.ai/","title":"Welo Data homepage","publisher":"Welo Data (Welocalize)","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T5","locator":"Offerings: data collection, annotation, multilingual RL environments, model evaluation and red teaming; 500k+ curated experts; case studies up to 2026-05-14; no revenue"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Method: Welo Data is Welocalize's AI-data brand (500k+ experts, multilingual RL environments, evaluation); no revenue is published and group figures are not used. Central US$80M is an estimate (US$30-220M). Unit only. Full pass: searches found no unit revenue figure. Main input: Welo Data (Welocalize) (T2). Share = company ÷ function total; low and high move only this company's own range and hold the others at central. Welo Data RL environments and agent tasks: US$8m (1.5-33m) = run-rate US$80m (30-220m) × 10% (the smaller parts of a mixed vendor are varied by ±50% and the largest part absorbs the difference). Function total, annualised US$m (central): 1,839, made of identified candidates 1,739 and an allowance of 100 for vendors outside the register (for example Deccan AI, Shaip, Cogito Tech, Sigma AI, Centaur Labs, Revelo, e2f, Patronus AI, Haize Labs, Lightly, Mostly AI, Abundant, Ethos, Datumo and smaller environment builders). Largest, US$m: mercor 484, surge-ai 414, proximal 140, handshake-ai 136, micro1 130, other candidates 434, vendors outside the register 100. Full table: a local evidence capture. Share 0.435% (range 0.082-1.771%). Source published date: undated (undated)."},"evaluation":{"low":0.129,"central":0.683,"high":2.757,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"Mid-2026 (estimate)","periodEnd":"2026-09-30","source":{"url":"https://welodata.ai/","title":"Welo Data homepage","publisher":"Welo Data (Welocalize)","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T5","locator":"Offerings: data collection, annotation, multilingual RL environments, model evaluation and red teaming; 500k+ curated experts; case studies up to 2026-05-14; no revenue"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Method: Welo Data is Welocalize's AI-data brand (500k+ experts, multilingual RL environments, evaluation); no revenue is published and group figures are not used. Central US$80M is an estimate (US$30-220M). Unit only. Full pass: searches found no unit revenue figure. Main input: Welo Data (Welocalize) (T2). Share = company ÷ function total; low and high move only this company's own range and hold the others at central. Welo Data independent model evaluation: US$4m (0.75-16.5m) = run-rate US$80m (30-220m) × 5% (the smaller parts of a mixed vendor are varied by ±50% and the largest part absorbs the difference). Function total, annualised US$m (central): 586, made of identified candidates 566 and an allowance of 20 for vendors outside the register (for example Deccan AI, Shaip, Cogito Tech, Sigma AI, Centaur Labs, Revelo, e2f, Patronus AI, Haize Labs, Lightly, Mostly AI, Abundant, Ethos, Datumo and smaller environment builders). Largest, US$m: arena-ai 110, surge-ai 90, mercor 66, scale-ai 45, innodata 33, other candidates 222, vendors outside the register 20. Full table: a local evidence capture. Share 0.683% (range 0.129-2.757%). Source published date: undated (undated)."},"data-tooling":{"participates":false,"evidence":{"url":"https://welodata.ai/","title":"Welo Data homepage","publisher":"Welo Data (Welocalize)","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T2","locator":"Offerings: data collection, annotation, multilingual RL environments, model evaluation and red teaming; 500k+ curated experts; case studies up to 2026-05-14; no revenue; no labelling and data software offering described"}}},"screening":{"expert-data":{"levelLow":1,"levelHigh":2},"labelling-multimodal":{"levelLow":1,"levelHigh":2},"rl-environments":{"levelLow":1,"levelHigh":2},"evaluation":{"levelLow":1,"levelHigh":2}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"3e14516ccc1b88c70872212825e1a201ea1d758bf4e4ab9792164bba9ca0892a"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"3e14516ccc1b88c70872212825e1a201ea1d758bf4e4ab9792164bba9ca0892a"}}},"imerit":{"schema":"5lc-index-inputs/1","key":"models-data-evaluation","entityId":"imerit","name":"iMerit","kind":"unit","parentId":"exl","commercialActivityVerified":true,"shares":{"expert-data":{"low":0.124,"central":0.356,"high":0.811,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"Aug-Dec 2026 guidance, annualised","periodEnd":"2026-12-31","source":{"url":"https://sec.gov/Archives/edgar/data/1297989/000129798926000021/exlq226pressrelease.htm","title":"EXL reports 2026 second quarter results (8-K exhibit 99.1)","publisher":"ExlService Holdings (SEC Form 8-K)","publishedDate":"2026-07-28","accessedDate":"2026-10-22","tier":"T5","locator":"The company's 2026 guidance includes $28.0M to $32.0M of anticipated revenue from the iMerit acquisition, expected to close on July 31, 2026; iMerit standalone revenue not disclosed"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Method: EXL's guidance includes US$28.0M to US$32.0M of anticipated revenue from iMerit; closing was expected on 31 July 2026, so the period is five months: US$30M x 12/5 = US$72M annualised (low US$50M for seasonality, high US$110M). Split follows the card notes (expert, labelling, Ango Hub software). Main input: ExlService Holdings (SEC Form 8-K) (T2). Share = company ÷ function total; low and high move only this company's own range and hold the others at central. iMerit expert data for post-training: US$25.2m (8.75-57.7m) = run-rate US$72m (50-110m) × 35% (the smaller parts of a mixed vendor are varied by ±50% and the largest part absorbs the difference). Function total, annualised US$m (central): 7,084, made of identified candidates 6,934 and an allowance of 150 for vendors outside the register (for example Deccan AI, Shaip, Cogito Tech, Sigma AI, Centaur Labs, Revelo, e2f, Patronus AI, Haize Labs, Lightly, Mostly AI, Abundant, Ethos, Datumo and smaller environment builders). Largest, US$m: mercor 1,650, surge-ai 1,296, handshake-ai 882, micro1 850, scale-ai 412, other candidates 1,844, vendors outside the register 150. Full table: a local evidence capture. Share 0.356% (range 0.124-0.811%). Source published date: 2026-07-28 (full date)."},"labelling-multimodal":{"low":0.72,"central":2.046,"high":4.569,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"Aug-Dec 2026 guidance, annualised","periodEnd":"2026-12-31","source":{"url":"https://sec.gov/Archives/edgar/data/1297989/000129798926000021/exlq226pressrelease.htm","title":"EXL reports 2026 second quarter results (8-K exhibit 99.1)","publisher":"ExlService Holdings (SEC Form 8-K)","publishedDate":"2026-07-28","accessedDate":"2026-10-22","tier":"T5","locator":"The company's 2026 guidance includes $28.0M to $32.0M of anticipated revenue from the iMerit acquisition, expected to close on July 31, 2026; iMerit standalone revenue not disclosed"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Method: EXL's guidance includes US$28.0M to US$32.0M of anticipated revenue from iMerit; closing was expected on 31 July 2026, so the period is five months: US$30M x 12/5 = US$72M annualised (low US$50M for seasonality, high US$110M). Split follows the card notes (expert, labelling, Ango Hub software). Main input: ExlService Holdings (SEC Form 8-K) (T2). Share = company ÷ function total; low and high move only this company's own range and hold the others at central. iMerit large-scale labelling and multimodal data: US$36m (12.5-82.5m) = run-rate US$72m (50-110m) × 50% (the smaller parts of a mixed vendor are varied by ±50% and the largest part absorbs the difference). Function total, annualised US$m (central): 1,759, made of identified candidates 1,609 and an allowance of 150 for vendors outside the register (for example Deccan AI, Shaip, Cogito Tech, Sigma AI, Centaur Labs, Revelo, e2f, Patronus AI, Haize Labs, Lightly, Mostly AI, Abundant, Ethos, Datumo and smaller environment builders). Largest, US$m: scale-ai 262, appen 176, taskus 146, rws-trainai 140, telus-digital 124, other candidates 762, vendors outside the register 150. Full table: a local evidence capture. Share 2.046% (range 0.72-4.569%). Source published date: 2026-07-28 (full date)."},"rl-environments":{"participates":false,"evidence":{"url":"https://sec.gov/Archives/edgar/data/1297989/000129798926000021/exlq226pressrelease.htm","title":"EXL reports 2026 second quarter results (8-K exhibit 99.1)","publisher":"ExlService Holdings (SEC Form 8-K)","publishedDate":"2026-07-28","accessedDate":"2026-10-22","tier":"T2","locator":"The company's 2026 guidance includes $28.0M to $32.0M of anticipated revenue from the iMerit acquisition, expected to close on July 31, 2026; iMerit standalone revenue not disclosed; no RL environments and agent tasks offering described"}},"evaluation":{"participates":false,"evidence":{"url":"https://sec.gov/Archives/edgar/data/1297989/000129798926000021/exlq226pressrelease.htm","title":"EXL reports 2026 second quarter results (8-K exhibit 99.1)","publisher":"ExlService Holdings (SEC Form 8-K)","publishedDate":"2026-07-28","accessedDate":"2026-10-22","tier":"T2","locator":"The company's 2026 guidance includes $28.0M to $32.0M of anticipated revenue from the iMerit acquisition, expected to close on July 31, 2026; iMerit standalone revenue not disclosed; no independent model evaluation offering described"}},"data-tooling":{"low":1.461,"central":4.096,"high":8.916,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"Aug-Dec 2026 guidance, annualised","periodEnd":"2026-12-31","source":{"url":"https://sec.gov/Archives/edgar/data/1297989/000129798926000021/exlq226pressrelease.htm","title":"EXL reports 2026 second quarter results (8-K exhibit 99.1)","publisher":"ExlService Holdings (SEC Form 8-K)","publishedDate":"2026-07-28","accessedDate":"2026-10-22","tier":"T5","locator":"The company's 2026 guidance includes $28.0M to $32.0M of anticipated revenue from the iMerit acquisition, expected to close on July 31, 2026; iMerit standalone revenue not disclosed"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Method: EXL's guidance includes US$28.0M to US$32.0M of anticipated revenue from iMerit; closing was expected on 31 July 2026, so the period is five months: US$30M x 12/5 = US$72M annualised (low US$50M for seasonality, high US$110M). Split follows the card notes (expert, labelling, Ango Hub software). Main input: ExlService Holdings (SEC Form 8-K) (T2). Share = company ÷ function total; low and high move only this company's own range and hold the others at central. iMerit labelling and data software: US$10.8m (3.75-24.7m) = run-rate US$72m (50-110m) × 15% (the smaller parts of a mixed vendor are varied by ±50% and the largest part absorbs the difference). Function total, annualised US$m (central): 264, made of identified candidates 234 and an allowance of 30 for vendors outside the register (for example Deccan AI, Shaip, Cogito Tech, Sigma AI, Centaur Labs, Revelo, e2f, Patronus AI, Haize Labs, Lightly, Mostly AI, Abundant, Ethos, Datumo and smaller environment builders). Largest, US$m: snorkel-ai 37.5, labelbox 37.5, encord 36, roboflow 25, superannotate 21, other candidates 76.7, vendors outside the register 30. Full table: a local evidence capture. Share 4.096% (range 1.461-8.916%). Source published date: 2026-07-28 (full date)."}},"screening":{"expert-data":{"levelLow":1,"levelHigh":2},"labelling-multimodal":{"levelLow":1,"levelHigh":2},"data-tooling":{"levelLow":1,"levelHigh":2}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"843f8d3c432db96f6061b393e639ee60e566553910e484beaca6ad6c5d176f6c"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"843f8d3c432db96f6061b393e639ee60e566553910e484beaca6ad6c5d176f6c"}}},"lionbridge":{"schema":"5lc-index-inputs/1","key":"models-data-evaluation","entityId":"lionbridge","name":"Lionbridge (AI training services)","kind":"private","commercialActivityVerified":false,"shares":{"expert-data":{"low":0.035,"central":0.247,"high":1.05,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"Mid-2026 (estimate)","periodEnd":"2026-09-30","source":{"url":"https://www.lionbridge.com/","title":"Lionbridge homepage","publisher":"Lionbridge","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T5","locator":"AI Data Services listed (Aurora AI Studio); half a million experts; case studies anonymised; no AI revenue figure"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Method: No AI-data revenue is published; the site lists AI Data Services, half a million experts and anonymised case studies. Central US$70M is an estimate (US$20-200M).. Full pass: searches found no AI-data revenue figure. Main input: Lionbridge (T2). Share = company ÷ function total; low and high move only this company's own range and hold the others at central. Lionbridge (AI training services) expert data for post-training: US$17.5m (2.5-75m) = run-rate US$70m (20-200m) × 25% (the smaller parts of a mixed vendor are varied by ±50% and the largest part absorbs the difference). Function total, annualised US$m (central): 7,084, made of identified candidates 6,934 and an allowance of 150 for vendors outside the register (for example Deccan AI, Shaip, Cogito Tech, Sigma AI, Centaur Labs, Revelo, e2f, Patronus AI, Haize Labs, Lightly, Mostly AI, Abundant, Ethos, Datumo and smaller environment builders). Largest, US$m: mercor 1,650, surge-ai 1,296, handshake-ai 882, micro1 850, scale-ai 412, other candidates 1,844, vendors outside the register 150. Full table: a local evidence capture. Share 0.247% (range 0.035-1.05%). Source published date: undated (undated)."},"labelling-multimodal":{"low":0.639,"central":2.785,"high":9.041,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"Mid-2026 (estimate)","periodEnd":"2026-09-30","source":{"url":"https://www.lionbridge.com/","title":"Lionbridge homepage","publisher":"Lionbridge","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T5","locator":"AI Data Services listed (Aurora AI Studio); half a million experts; case studies anonymised; no AI revenue figure"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Method: No AI-data revenue is published; the site lists AI Data Services, half a million experts and anonymised case studies. Central US$70M is an estimate (US$20-200M).. Full pass: searches found no AI-data revenue figure. Main input: Lionbridge (T2). Share = company ÷ function total; low and high move only this company's own range and hold the others at central. Lionbridge (AI training services) large-scale labelling and multimodal data: US$49m (11-170m) = run-rate US$70m (20-200m) × 70% (the smaller parts of a mixed vendor are varied by ±50% and the largest part absorbs the difference). Function total, annualised US$m (central): 1,759, made of identified candidates 1,609 and an allowance of 150 for vendors outside the register (for example Deccan AI, Shaip, Cogito Tech, Sigma AI, Centaur Labs, Revelo, e2f, Patronus AI, Haize Labs, Lightly, Mostly AI, Abundant, Ethos, Datumo and smaller environment builders). Largest, US$m: scale-ai 262, appen 176, taskus 146, rws-trainai 140, telus-digital 124, other candidates 762, vendors outside the register 150. Full table: a local evidence capture. Share 2.785% (range 0.639-9.041%). Source published date: undated (undated)."},"rl-environments":{"participates":false,"evidence":{"url":"https://www.lionbridge.com/","title":"Lionbridge homepage","publisher":"Lionbridge","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T2","locator":"AI Data Services listed (Aurora AI Studio); half a million experts; case studies anonymised; no AI revenue figure; no RL environments and agent tasks offering described"}},"evaluation":{"low":0.086,"central":0.597,"high":2.511,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"Mid-2026 (estimate)","periodEnd":"2026-09-30","source":{"url":"https://www.lionbridge.com/","title":"Lionbridge homepage","publisher":"Lionbridge","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T5","locator":"AI Data Services listed (Aurora AI Studio); half a million experts; case studies anonymised; no AI revenue figure"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Method: No AI-data revenue is published; the site lists AI Data Services, half a million experts and anonymised case studies. Central US$70M is an estimate (US$20-200M).. Full pass: searches found no AI-data revenue figure. Main input: Lionbridge (T2). Share = company ÷ function total; low and high move only this company's own range and hold the others at central. Lionbridge (AI training services) independent model evaluation: US$3.5m (0.5-15m) = run-rate US$70m (20-200m) × 5% (the smaller parts of a mixed vendor are varied by ±50% and the largest part absorbs the difference). Function total, annualised US$m (central): 586, made of identified candidates 566 and an allowance of 20 for vendors outside the register (for example Deccan AI, Shaip, Cogito Tech, Sigma AI, Centaur Labs, Revelo, e2f, Patronus AI, Haize Labs, Lightly, Mostly AI, Abundant, Ethos, Datumo and smaller environment builders). Largest, US$m: arena-ai 110, surge-ai 90, mercor 66, scale-ai 45, innodata 33, other candidates 222, vendors outside the register 20. Full table: a local evidence capture. Share 0.597% (range 0.086-2.511%). Source published date: undated (undated)."},"data-tooling":{"participates":false,"evidence":{"url":"https://www.lionbridge.com/","title":"Lionbridge homepage","publisher":"Lionbridge","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T2","locator":"AI Data Services listed (Aurora AI Studio); half a million experts; case studies anonymised; no AI revenue figure; no labelling and data software offering described"}}},"screening":{"expert-data":{"levelLow":1,"levelHigh":2},"labelling-multimodal":{"levelLow":1,"levelHigh":2},"evaluation":{"levelLow":1,"levelHigh":2}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"7303fedd547bfd25ff958eb1405ee18c923887fca02b164c9863a338d6bfb6c6"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"7303fedd547bfd25ff958eb1405ee18c923887fca02b164c9863a338d6bfb6c6"}}},"lxt":{"schema":"5lc-index-inputs/1","key":"models-data-evaluation","entityId":"lxt","name":"LXT","kind":"private","commercialActivityVerified":false,"shares":{"expert-data":{"low":0.035,"central":0.198,"high":0.758,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"Mid-2026 (estimate)","periodEnd":"2026-09-30","source":{"url":"https://www.lxt.ai/","title":"LXT homepage","publisher":"LXT","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T5","locator":"AI training data services (collection, annotation, RLHF, red-teaming, datasets); 10M+ contributors; press release dated 2026-06-15; no revenue"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Method: No revenue is published. LXT (with clickworker) says it serves the top ten global technology companies, with 10M+ contributors and 250K+ domain experts; no customers are named. Central US$70M is an estimate (US$25-180M). Full pass: no new figure found. Independent evaluation (added at the check-fix pass of 22 October 2026): the homepage lists AI model evaluation, LLM red-teaming, human-led evaluation and prompt evaluation next to collection and annotation, with no revenue split; 10% of the run-rate is taken from labelling (80% to 70%) as a judgment. Main input: LXT (T2). Share = company ÷ function total; low and high move only this company's own range and hold the others at central. LXT expert data for post-training: US$14m (2.5-54m) = run-rate US$70m (25-180m) × 20% (the smaller parts of a mixed vendor are varied by ±50% and the largest part absorbs the difference). Function total, annualised US$m (central): 7,084, made of identified candidates 6,934 and an allowance of 150 for vendors outside the register (for example Deccan AI, Shaip, Cogito Tech, Sigma AI, Centaur Labs, Revelo, e2f, Patronus AI, Haize Labs, Lightly, Mostly AI, Abundant, Ethos, Datumo and smaller environment builders). Largest, US$m: mercor 1,650, surge-ai 1,296, handshake-ai 882, micro1 850, scale-ai 412, other candidates 1,844, vendors outside the register 150. Full table: a local evidence capture. Share 0.198% (range 0.035-0.758%). Source published date: undated (undated)."},"labelling-multimodal":{"low":0.798,"central":2.785,"high":8.211,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"Mid-2026 (estimate)","periodEnd":"2026-09-30","source":{"url":"https://www.lxt.ai/","title":"LXT homepage","publisher":"LXT","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T5","locator":"AI training data services (collection, annotation, RLHF, red-teaming, datasets); 10M+ contributors; press release dated 2026-06-15; no revenue"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Method: No revenue is published. LXT (with clickworker) says it serves the top ten global technology companies, with 10M+ contributors and 250K+ domain experts; no customers are named. Central US$70M is an estimate (US$25-180M). Full pass: no new figure found. Independent evaluation (added at the check-fix pass of 22 October 2026): the homepage lists AI model evaluation, LLM red-teaming, human-led evaluation and prompt evaluation next to collection and annotation, with no revenue split; 10% of the run-rate is taken from labelling (80% to 70%) as a judgment. Main input: LXT (T2). Share = company ÷ function total; low and high move only this company's own range and hold the others at central. LXT large-scale labelling and multimodal data: US$49m (13.7-153m) = run-rate US$70m (25-180m) × 70% (the smaller parts of a mixed vendor are varied by ±50% and the largest part absorbs the difference). Function total, annualised US$m (central): 1,759, made of identified candidates 1,609 and an allowance of 150 for vendors outside the register (for example Deccan AI, Shaip, Cogito Tech, Sigma AI, Centaur Labs, Revelo, e2f, Patronus AI, Haize Labs, Lightly, Mostly AI, Abundant, Ethos, Datumo and smaller environment builders). Largest, US$m: scale-ai 262, appen 176, taskus 146, rws-trainai 140, telus-digital 124, other candidates 762, vendors outside the register 150. Full table: a local evidence capture. Share 2.785% (range 0.798-8.211%). Source published date: undated (undated)."},"rl-environments":{"participates":false,"evidence":{"url":"https://www.lxt.ai/","title":"LXT homepage","publisher":"LXT","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T2","locator":"AI training data services (collection, annotation, RLHF, red-teaming, datasets); 10M+ contributors; press release dated 2026-06-15; no revenue; no RL environments and agent tasks offering described"}},"evaluation":{"low":0.215,"central":1.195,"high":4.456,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"Mid-2026 (estimate)","periodEnd":"2026-09-30","source":{"url":"https://www.lxt.ai/","title":"LXT homepage","publisher":"LXT","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T5","locator":"Data evaluation: 'AI model evaluation' ('evaluating model outputs post-deployment'), 'LLM red-teaming' ('We run red teaming, safety tests, and expert reviews to surface bias or risk'), 'Human-led evaluation', 'Prompt Engineering & Evaluation'; no revenue"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Method: No revenue is published. LXT (with clickworker) says it serves the top ten global technology companies, with 10M+ contributors and 250K+ domain experts; no customers are named. Central US$70M is an estimate (US$25-180M). Full pass: no new figure found. Independent evaluation (added at the check-fix pass of 22 October 2026): the homepage lists AI model evaluation, LLM red-teaming, human-led evaluation and prompt evaluation next to collection and annotation, with no revenue split; 10% of the run-rate is taken from labelling (80% to 70%) as a judgment. Main input: LXT (T2). Share = company ÷ function total; low and high move only this company's own range and hold the others at central. LXT independent model evaluation: US$7m (1.25-27m) = run-rate US$70m (25-180m) × 10% (the smaller parts of a mixed vendor are varied by ±50% and the largest part absorbs the difference). Function total, annualised US$m (central): 586, made of identified candidates 566 and an allowance of 20 for vendors outside the register (for example Deccan AI, Shaip, Cogito Tech, Sigma AI, Centaur Labs, Revelo, e2f, Patronus AI, Haize Labs, Lightly, Mostly AI, Abundant, Ethos, Datumo and smaller environment builders). Largest, US$m: arena-ai 110, surge-ai 90, mercor 66, scale-ai 45, innodata 33, other candidates 222, vendors outside the register 20. Full table: a local evidence capture. Share 1.195% (range 0.215-4.456%). Source published date: undated (undated)."},"data-tooling":{"participates":false,"evidence":{"url":"https://www.lxt.ai/","title":"LXT homepage","publisher":"LXT","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T2","locator":"AI training data services (collection, annotation, RLHF, red-teaming, datasets); 10M+ contributors; press release dated 2026-06-15; no revenue; no labelling and data software offering described"}}},"screening":{"expert-data":{"levelLow":1,"levelHigh":2},"labelling-multimodal":{"levelLow":1,"levelHigh":2},"evaluation":{"levelLow":1,"levelHigh":2}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"63b6aea818b6d2994454c1549ac526fd335b071bf5e2674d1291c60a92df3a90"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"63b6aea818b6d2994454c1549ac526fd335b071bf5e2674d1291c60a92df3a90"}}},"dataocean-ai":{"schema":"5lc-index-inputs/1","key":"models-data-evaluation","entityId":"dataocean-ai","name":"DataOcean AI","kind":"public","commercialActivityVerified":true,"shares":{"expert-data":{"low":0.016,"central":0.038,"high":0.069,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"H1 2026, annualised","periodEnd":"2026-06-30","source":{"url":"https://finance.sina.cn/stock/jdts/2026-10-09/detail-iniurwaf7778162.d.html","title":"Haitian Ruisheng (DataOcean AI) H1 2026 results coverage","publisher":"Sina Finance","publishedDate":"2026-10-09","accessedDate":"2026-10-22","tier":"T5","locator":"H1 2026 revenue RMB 192M (+22.45%), Q2 RMB 95M; suppliers to Zhipu, Meta, Alibaba, Tencent, Moonshot and MiniMax"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Method: Haitian Ruisheng (DataOcean AI) H1 2026 revenue RMB 192M (Q2 RMB 95M): annualised RMB 384M at an assumed RMB 7.1 per US dollar = US$54M (low US$45M with the exchange rate and seasonality, high US$65M). Chinese-language speech and vision data for Chinese and US model builders (Zhipu, Meta, Alibaba, Tencent). Split is an estimate (mostly speech and vision data collection). Main input: Sina Finance (T3). Share = company ÷ function total; low and high move only this company's own range and hold the others at central. DataOcean AI expert data for post-training: US$2.7m (1.12-4.88m) = run-rate US$54m (45-65m) × 5% (the smaller parts of a mixed vendor are varied by ±50% and the largest part absorbs the difference). Function total, annualised US$m (central): 7,084, made of identified candidates 6,934 and an allowance of 150 for vendors outside the register (for example Deccan AI, Shaip, Cogito Tech, Sigma AI, Centaur Labs, Revelo, e2f, Patronus AI, Haize Labs, Lightly, Mostly AI, Abundant, Ethos, Datumo and smaller environment builders). Largest, US$m: mercor 1,650, surge-ai 1,296, handshake-ai 882, micro1 850, scale-ai 412, other candidates 1,844, vendors outside the register 150. Full table: a local evidence capture. Share 0.038% (range 0.016-0.069%). Source published date: 2026-10-09 (full date)."},"labelling-multimodal":{"low":2.379,"central":2.916,"high":3.578,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"H1 2026, annualised","periodEnd":"2026-06-30","source":{"url":"https://finance.sina.cn/stock/jdts/2026-10-09/detail-iniurwaf7778162.d.html","title":"Haitian Ruisheng (DataOcean AI) H1 2026 results coverage","publisher":"Sina Finance","publishedDate":"2026-10-09","accessedDate":"2026-10-22","tier":"T5","locator":"H1 2026 revenue RMB 192M (+22.45%), Q2 RMB 95M; suppliers to Zhipu, Meta, Alibaba, Tencent, Moonshot and MiniMax"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Method: Haitian Ruisheng (DataOcean AI) H1 2026 revenue RMB 192M (Q2 RMB 95M): annualised RMB 384M at an assumed RMB 7.1 per US dollar = US$54M (low US$45M with the exchange rate and seasonality, high US$65M). Chinese-language speech and vision data for Chinese and US model builders (Zhipu, Meta, Alibaba, Tencent). Split is an estimate (mostly speech and vision data collection). Main input: Sina Finance (T3). Share = company ÷ function total; low and high move only this company's own range and hold the others at central. DataOcean AI large-scale labelling and multimodal data: US$51.3m (41.6-63.4m) = run-rate US$54m (45-65m) × 95% (the smaller parts of a mixed vendor are varied by ±50% and the largest part absorbs the difference). Function total, annualised US$m (central): 1,759, made of identified candidates 1,609 and an allowance of 150 for vendors outside the register (for example Deccan AI, Shaip, Cogito Tech, Sigma AI, Centaur Labs, Revelo, e2f, Patronus AI, Haize Labs, Lightly, Mostly AI, Abundant, Ethos, Datumo and smaller environment builders). Largest, US$m: scale-ai 262, appen 176, taskus 146, rws-trainai 140, telus-digital 124, other candidates 762, vendors outside the register 150. Full table: a local evidence capture. Share 2.916% (range 2.379-3.578%). Source published date: 2026-10-09 (full date)."},"rl-environments":{"participates":false,"evidence":{"url":"https://finance.sina.cn/stock/jdts/2026-10-09/detail-iniurwaf7778162.d.html","title":"Haitian Ruisheng (DataOcean AI) H1 2026 results coverage","publisher":"Sina Finance","publishedDate":"2026-10-09","accessedDate":"2026-10-22","tier":"T3","locator":"H1 2026 revenue RMB 192M (+22.45%), Q2 RMB 95M; suppliers to Zhipu, Meta, Alibaba, Tencent, Moonshot and MiniMax; no RL environments and agent tasks offering described"}},"evaluation":{"participates":false,"evidence":{"url":"https://finance.sina.cn/stock/jdts/2026-10-09/detail-iniurwaf7778162.d.html","title":"Haitian Ruisheng (DataOcean AI) H1 2026 results coverage","publisher":"Sina Finance","publishedDate":"2026-10-09","accessedDate":"2026-10-22","tier":"T3","locator":"H1 2026 revenue RMB 192M (+22.45%), Q2 RMB 95M; suppliers to Zhipu, Meta, Alibaba, Tencent, Moonshot and MiniMax; no independent model evaluation offering described"}},"data-tooling":{"participates":false,"evidence":{"url":"https://finance.sina.cn/stock/jdts/2026-10-09/detail-iniurwaf7778162.d.html","title":"Haitian Ruisheng (DataOcean AI) H1 2026 results coverage","publisher":"Sina Finance","publishedDate":"2026-10-09","accessedDate":"2026-10-22","tier":"T3","locator":"H1 2026 revenue RMB 192M (+22.45%), Q2 RMB 95M; suppliers to Zhipu, Meta, Alibaba, Tencent, Moonshot and MiniMax; no labelling and data software offering described"}}},"screening":{"expert-data":{"levelLow":1,"levelHigh":2},"labelling-multimodal":{"levelLow":1,"levelHigh":3}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"3dee78a83802a222846799c0f93e0d2136023f2cb02e378f84bd655376b2fd32"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"3dee78a83802a222846799c0f93e0d2136023f2cb02e378f84bd655376b2fd32"}}},"encord":{"schema":"5lc-index-inputs/1","key":"models-data-evaluation","entityId":"encord","name":"Encord","kind":"private","commercialActivityVerified":true,"shares":{"expert-data":{"participates":false,"evidence":{"url":"https://www.bolsamania.com/nota-de-prensa_amp/mercados/c-o-r-r-e-c-t-i-o-n----encord--21864076.html","title":"Encord Series C press release (corrected)","publisher":"Encord (PR Newswire, via Bolsamania)","publishedDate":"2026-02-26","accessedDate":"2026-10-22","tier":"T2","locator":"The company says: $60M Series C, revenue from physical AI customers up 10x in twelve months, over 300 AI teams; customers include Woven by Toyota, Skydio, AXA; no revenue figure; no expert data for post-training offering described"}},"labelling-multimodal":{"low":0.057,"central":0.227,"high":0.679,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"Early 2026 (estimate)","periodEnd":"2026-02-26","source":{"url":"https://www.bolsamania.com/nota-de-prensa_amp/mercados/c-o-r-r-e-c-t-i-o-n----encord--21864076.html","title":"Encord Series C press release (corrected)","publisher":"Encord (PR Newswire, via Bolsamania)","publishedDate":"2026-02-26","accessedDate":"2026-10-22","tier":"T5","locator":"The company says: $60M Series C, revenue from physical AI customers up 10x in twelve months, over 300 AI teams; customers include Woven by Toyota, Skydio, AXA; no revenue figure"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Method: The company says (26 Feb 2026) revenue from physical-AI customers grew 10x in twelve months, with 300+ AI teams and customers such as Woven by Toyota, Skydio and AXA; no revenue figure (third-party databases US$1-46M, not used). Central US$40M is an estimate (US$20-80M); 10% labelling services. Main input: Encord (PR Newswire, via Bolsamania) (T2). Share = company ÷ function total; low and high move only this company's own range and hold the others at central. Encord large-scale labelling and multimodal data: US$4m (1-12m) = run-rate US$40m (20-80m) × 10% (the smaller parts of a mixed vendor are varied by ±50% and the largest part absorbs the difference). Function total, annualised US$m (central): 1,759, made of identified candidates 1,609 and an allowance of 150 for vendors outside the register (for example Deccan AI, Shaip, Cogito Tech, Sigma AI, Centaur Labs, Revelo, e2f, Patronus AI, Haize Labs, Lightly, Mostly AI, Abundant, Ethos, Datumo and smaller environment builders). Largest, US$m: scale-ai 262, appen 176, taskus 146, rws-trainai 140, telus-digital 124, other candidates 762, vendors outside the register 150. Full table: a local evidence capture. Share 0.227% (range 0.057-0.679%). Source published date: 2026-02-26 (full date)."},"rl-environments":{"participates":false,"evidence":{"url":"https://www.bolsamania.com/nota-de-prensa_amp/mercados/c-o-r-r-e-c-t-i-o-n----encord--21864076.html","title":"Encord Series C press release (corrected)","publisher":"Encord (PR Newswire, via Bolsamania)","publishedDate":"2026-02-26","accessedDate":"2026-10-22","tier":"T2","locator":"The company says: $60M Series C, revenue from physical AI customers up 10x in twelve months, over 300 AI teams; customers include Woven by Toyota, Skydio, AXA; no revenue figure; no RL environments and agent tasks offering described"}},"evaluation":{"participates":false,"evidence":{"url":"https://www.bolsamania.com/nota-de-prensa_amp/mercados/c-o-r-r-e-c-t-i-o-n----encord--21864076.html","title":"Encord Series C press release (corrected)","publisher":"Encord (PR Newswire, via Bolsamania)","publishedDate":"2026-02-26","accessedDate":"2026-10-22","tier":"T2","locator":"The company says: $60M Series C, revenue from physical AI customers up 10x in twelve months, over 300 AI teams; customers include Woven by Toyota, Skydio, AXA; no revenue figure; no independent model evaluation offering described"}},"data-tooling":{"low":6.949,"central":13.654,"high":25.029,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"Early 2026 (estimate)","periodEnd":"2026-02-26","source":{"url":"https://www.bolsamania.com/nota-de-prensa_amp/mercados/c-o-r-r-e-c-t-i-o-n----encord--21864076.html","title":"Encord Series C press release (corrected)","publisher":"Encord (PR Newswire, via Bolsamania)","publishedDate":"2026-02-26","accessedDate":"2026-10-22","tier":"T5","locator":"The company says: $60M Series C, revenue from physical AI customers up 10x in twelve months, over 300 AI teams; customers include Woven by Toyota, Skydio, AXA; no revenue figure"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Method: The company says (26 Feb 2026) revenue from physical-AI customers grew 10x in twelve months, with 300+ AI teams and customers such as Woven by Toyota, Skydio and AXA; no revenue figure (third-party databases US$1-46M, not used). Central US$40M is an estimate (US$20-80M); 10% labelling services. Main input: Encord (PR Newswire, via Bolsamania) (T2). Share = company ÷ function total; low and high move only this company's own range and hold the others at central. Encord labelling and data software: US$36m (17-76m) = run-rate US$40m (20-80m) × 90% (the smaller parts of a mixed vendor are varied by ±50% and the largest part absorbs the difference). Function total, annualised US$m (central): 264, made of identified candidates 234 and an allowance of 30 for vendors outside the register (for example Deccan AI, Shaip, Cogito Tech, Sigma AI, Centaur Labs, Revelo, e2f, Patronus AI, Haize Labs, Lightly, Mostly AI, Abundant, Ethos, Datumo and smaller environment builders). Largest, US$m: snorkel-ai 37.5, labelbox 37.5, encord 36, roboflow 25, superannotate 21, other candidates 76.7, vendors outside the register 30. Full table: a local evidence capture. Share 13.654% (range 6.949-25.029%). Source published date: 2026-02-26 (full date)."}},"screening":{"labelling-multimodal":{"levelLow":1,"levelHigh":2},"data-tooling":{"levelLow":1,"levelHigh":2}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"0d8c356252c1c6006bf79365580499d33e96454ccd3cb1dcbb0718f1d56c2de0"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"0d8c356252c1c6006bf79365580499d33e96454ccd3cb1dcbb0718f1d56c2de0"}}},"datatang":{"schema":"5lc-index-inputs/1","key":"models-data-evaluation","entityId":"datatang","name":"Datatang","kind":"public","commercialActivityVerified":false,"shares":{"expert-data":{"low":0.011,"central":0.056,"high":0.211,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"Mid-2026 (estimate)","periodEnd":"2026-09-30","source":{"url":"https://www.datatang.com/","title":"Datatang homepage","publisher":"Datatang","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T5","locator":"Offerings: AI training datasets, custom data collection, annotation platform; news items dated to 2026-10-14; no revenue or named customers"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Method: No revenue is stated on the page found. Datatang is a Chinese training-data company (founded 2010) selling datasets, custom collection and an annotation platform; the Chinese peer Haitian Ruisheng has H1 2026 revenue of about US$27M, so central US$40M annualised is an estimate with a wide range (US$15-100M). Main input: Datatang (T2). Share = company ÷ function total; low and high move only this company's own range and hold the others at central. Datatang expert data for post-training: US$4m (0.75-15m) = run-rate US$40m (15-100m) × 10% (the smaller parts of a mixed vendor are varied by ±50% and the largest part absorbs the difference). Function total, annualised US$m (central): 7,084, made of identified candidates 6,934 and an allowance of 150 for vendors outside the register (for example Deccan AI, Shaip, Cogito Tech, Sigma AI, Centaur Labs, Revelo, e2f, Patronus AI, Haize Labs, Lightly, Mostly AI, Abundant, Ethos, Datumo and smaller environment builders). Largest, US$m: mercor 1,650, surge-ai 1,296, handshake-ai 882, micro1 850, scale-ai 412, other candidates 1,844, vendors outside the register 150. Full table: a local evidence capture. Share 0.056% (range 0.011-0.211%). Source published date: undated (undated)."},"labelling-multimodal":{"low":0.669,"central":1.933,"high":5.089,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"Mid-2026 (estimate)","periodEnd":"2026-09-30","source":{"url":"https://www.datatang.com/","title":"Datatang homepage","publisher":"Datatang","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T5","locator":"Offerings: AI training datasets, custom data collection, annotation platform; news items dated to 2026-10-14; no revenue or named customers"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Method: No revenue is stated on the page found. Datatang is a Chinese training-data company (founded 2010) selling datasets, custom collection and an annotation platform; the Chinese peer Haitian Ruisheng has H1 2026 revenue of about US$27M, so central US$40M annualised is an estimate with a wide range (US$15-100M). Main input: Datatang (T2). Share = company ÷ function total; low and high move only this company's own range and hold the others at central. Datatang large-scale labelling and multimodal data: US$34m (11.6-92.5m) = run-rate US$40m (15-100m) × 85% (the smaller parts of a mixed vendor are varied by ±50% and the largest part absorbs the difference). Function total, annualised US$m (central): 1,759, made of identified candidates 1,609 and an allowance of 150 for vendors outside the register (for example Deccan AI, Shaip, Cogito Tech, Sigma AI, Centaur Labs, Revelo, e2f, Patronus AI, Haize Labs, Lightly, Mostly AI, Abundant, Ethos, Datumo and smaller environment builders). Largest, US$m: scale-ai 262, appen 176, taskus 146, rws-trainai 140, telus-digital 124, other candidates 762, vendors outside the register 150. Full table: a local evidence capture. Share 1.933% (range 0.669-5.089%). Source published date: undated (undated)."},"rl-environments":{"participates":false,"evidence":{"url":"https://www.datatang.com/","title":"Datatang homepage","publisher":"Datatang","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T2","locator":"Offerings: AI training datasets, custom data collection, annotation platform; news items dated to 2026-10-14; no revenue or named customers; no RL environments and agent tasks offering described"}},"evaluation":{"participates":false,"evidence":{"url":"https://www.datatang.com/","title":"Datatang homepage","publisher":"Datatang","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T2","locator":"Offerings: AI training datasets, custom data collection, annotation platform; news items dated to 2026-10-14; no revenue or named customers; no independent model evaluation offering described"}},"data-tooling":{"low":0.143,"central":0.759,"high":2.787,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"Mid-2026 (estimate)","periodEnd":"2026-09-30","source":{"url":"https://www.datatang.com/","title":"Datatang homepage","publisher":"Datatang","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T5","locator":"Offerings: AI training datasets, custom data collection, annotation platform; news items dated to 2026-10-14; no revenue or named customers"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Method: No revenue is stated on the page found. Datatang is a Chinese training-data company (founded 2010) selling datasets, custom collection and an annotation platform; the Chinese peer Haitian Ruisheng has H1 2026 revenue of about US$27M, so central US$40M annualised is an estimate with a wide range (US$15-100M). Main input: Datatang (T2). Share = company ÷ function total; low and high move only this company's own range and hold the others at central. Datatang labelling and data software: US$2m (0.38-7.5m) = run-rate US$40m (15-100m) × 5% (the smaller parts of a mixed vendor are varied by ±50% and the largest part absorbs the difference). Function total, annualised US$m (central): 264, made of identified candidates 234 and an allowance of 30 for vendors outside the register (for example Deccan AI, Shaip, Cogito Tech, Sigma AI, Centaur Labs, Revelo, e2f, Patronus AI, Haize Labs, Lightly, Mostly AI, Abundant, Ethos, Datumo and smaller environment builders). Largest, US$m: snorkel-ai 37.5, labelbox 37.5, encord 36, roboflow 25, superannotate 21, other candidates 76.7, vendors outside the register 30. Full table: a local evidence capture. Share 0.759% (range 0.143-2.787%). Source published date: undated (undated)."}},"screening":{"expert-data":{"levelLow":1,"levelHigh":2},"labelling-multimodal":{"levelLow":1,"levelHigh":2},"data-tooling":{"levelLow":1,"levelHigh":2}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"ac8ed122e724439e1df54a995ca28b879e5f82920033dbe1bfd85cb14fbbb4f3"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"ac8ed122e724439e1df54a995ca28b879e5f82920033dbe1bfd85cb14fbbb4f3"}}},"sama":{"schema":"5lc-index-inputs/1","key":"models-data-evaluation","entityId":"sama","name":"Sama","kind":"private","commercialActivityVerified":true,"shares":{"expert-data":{"low":0.011,"central":0.085,"high":0.38,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"Mid-2026 (estimate)","periodEnd":"2026-09-30","source":{"url":"https://www.sama.com/","title":"Sama homepage","publisher":"Sama","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T5","locator":"Offerings: image, video, 3D point cloud and text annotation, validation, model evaluation; case studies name Walmart; no revenue"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Method: No revenue is published. Sama sells image, video, 3D and text annotation to enterprises (Walmart case study; logos Getty, Microsoft, Qualcomm), mostly not to model developers; it stopped its content-moderation service in January 2023 and moved to AI training work (Business Daily Africa, Aug 2023, opened). Central US$40M is an estimate (US$10-120M). Full pass: searches found no revenue figure (aggregator estimates are not used). Independent evaluation (added at the check-fix pass of 22 October 2026): the homepage lists 'Model evaluation' (under Sama GenAI) and 'Sama Validate' next to annotation, with no revenue split; 5% of the run-rate is taken from labelling (85% to 80%) as a judgment. Red teaming and benchmarking are not mentioned. Main input: Sama (T2). Share = company ÷ function total; low and high move only this company's own range and hold the others at central. Sama expert data for post-training: US$6m (0.75-27m) = run-rate US$40m (10-120m) × 15% (the smaller parts of a mixed vendor are varied by ±50% and the largest part absorbs the difference). Function total, annualised US$m (central): 7,084, made of identified candidates 6,934 and an allowance of 150 for vendors outside the register (for example Deccan AI, Shaip, Cogito Tech, Sigma AI, Centaur Labs, Revelo, e2f, Patronus AI, Haize Labs, Lightly, Mostly AI, Abundant, Ethos, Datumo and smaller environment builders). Largest, US$m: mercor 1,650, surge-ai 1,296, handshake-ai 882, micro1 850, scale-ai 412, other candidates 1,844, vendors outside the register 150. Full table: a local evidence capture. Share 0.085% (range 0.011-0.38%). Source published date: undated (undated)."},"labelling-multimodal":{"low":0.404,"central":1.819,"high":5.885,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"Mid-2026 (estimate)","periodEnd":"2026-09-30","source":{"url":"https://www.sama.com/","title":"Sama homepage","publisher":"Sama","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T5","locator":"Offerings: image, video, 3D point cloud and text annotation, validation, model evaluation; case studies name Walmart; no revenue"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Method: No revenue is published. Sama sells image, video, 3D and text annotation to enterprises (Walmart case study; logos Getty, Microsoft, Qualcomm), mostly not to model developers; it stopped its content-moderation service in January 2023 and moved to AI training work (Business Daily Africa, Aug 2023, opened). Central US$40M is an estimate (US$10-120M). Full pass: searches found no revenue figure (aggregator estimates are not used). Independent evaluation (added at the check-fix pass of 22 October 2026): the homepage lists 'Model evaluation' (under Sama GenAI) and 'Sama Validate' next to annotation, with no revenue split; 5% of the run-rate is taken from labelling (85% to 80%) as a judgment. Red teaming and benchmarking are not mentioned. Main input: Sama (T2). Share = company ÷ function total; low and high move only this company's own range and hold the others at central. Sama large-scale labelling and multimodal data: US$32m (7-108m) = run-rate US$40m (10-120m) × 80% (the smaller parts of a mixed vendor are varied by ±50% and the largest part absorbs the difference). Function total, annualised US$m (central): 1,759, made of identified candidates 1,609 and an allowance of 150 for vendors outside the register (for example Deccan AI, Shaip, Cogito Tech, Sigma AI, Centaur Labs, Revelo, e2f, Patronus AI, Haize Labs, Lightly, Mostly AI, Abundant, Ethos, Datumo and smaller environment builders). Largest, US$m: scale-ai 262, appen 176, taskus 146, rws-trainai 140, telus-digital 124, other candidates 762, vendors outside the register 150. Full table: a local evidence capture. Share 1.819% (range 0.404-5.885%). Source published date: undated (undated)."},"rl-environments":{"participates":false,"evidence":{"url":"https://www.sama.com/","title":"Sama homepage","publisher":"Sama","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T2","locator":"Offerings: image, video, 3D point cloud and text annotation, validation, model evaluation; case studies name Walmart; no revenue; no RL environments and agent tasks offering described"}},"evaluation":{"low":0.043,"central":0.341,"high":1.518,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"Mid-2026 (estimate)","periodEnd":"2026-09-30","source":{"url":"https://www.sama.com/","title":"Sama homepage","publisher":"Sama","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T5","locator":"Sama GenAI 'Model Evaluation': 'Score real-world performance to guide retraining, iteration, and confident deployment'; 'Sama Validate' data validation; no revenue; red teaming and benchmarking are not mentioned"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Method: No revenue is published. Sama sells image, video, 3D and text annotation to enterprises (Walmart case study; logos Getty, Microsoft, Qualcomm), mostly not to model developers; it stopped its content-moderation service in January 2023 and moved to AI training work (Business Daily Africa, Aug 2023, opened). Central US$40M is an estimate (US$10-120M). Full pass: searches found no revenue figure (aggregator estimates are not used). Independent evaluation (added at the check-fix pass of 22 October 2026): the homepage lists 'Model evaluation' (under Sama GenAI) and 'Sama Validate' next to annotation, with no revenue split; 5% of the run-rate is taken from labelling (85% to 80%) as a judgment. Red teaming and benchmarking are not mentioned. Main input: Sama (T2). Share = company ÷ function total; low and high move only this company's own range and hold the others at central. Sama independent model evaluation: US$2m (0.25-9m) = run-rate US$40m (10-120m) × 5% (the smaller parts of a mixed vendor are varied by ±50% and the largest part absorbs the difference). Function total, annualised US$m (central): 586, made of identified candidates 566 and an allowance of 20 for vendors outside the register (for example Deccan AI, Shaip, Cogito Tech, Sigma AI, Centaur Labs, Revelo, e2f, Patronus AI, Haize Labs, Lightly, Mostly AI, Abundant, Ethos, Datumo and smaller environment builders). Largest, US$m: arena-ai 110, surge-ai 90, mercor 66, scale-ai 45, innodata 33, other candidates 222, vendors outside the register 20. Full table: a local evidence capture. Share 0.341% (range 0.043-1.518%). Source published date: undated (undated)."},"data-tooling":{"participates":false,"evidence":{"url":"https://www.sama.com/","title":"Sama homepage","publisher":"Sama","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T2","locator":"Offerings: image, video, 3D point cloud and text annotation, validation, model evaluation; case studies name Walmart; no revenue; no labelling and data software offering described"}}},"screening":{"expert-data":{"levelLow":1,"levelHigh":2},"labelling-multimodal":{"levelLow":1,"levelHigh":2},"evaluation":{"levelLow":1,"levelHigh":2}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"5235ccbb65df28a80bfd9416e89cff7a9e7e2622ab3cc18deeca06c80e4f7b0a"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"5235ccbb65df28a80bfd9416e89cff7a9e7e2622ab3cc18deeca06c80e4f7b0a"}}},"halluminate":{"schema":"5lc-index-inputs/1","key":"models-data-evaluation","entityId":"halluminate","name":"Halluminate","kind":"private","commercialActivityVerified":false,"shares":{"expert-data":{"participates":false,"evidence":{"url":"https://www.halluminate.ai/","title":"Halluminate homepage","publisher":"Halluminate","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T2","locator":"Data research lab building benchmarks and RL environments for finance and consulting workflows; Series A banner; benchmark dated 2026-08-11; no customers or revenue; no expert data for post-training offering described"}},"labelling-multimodal":{"participates":false,"evidence":{"url":"https://www.halluminate.ai/","title":"Halluminate homepage","publisher":"Halluminate","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T2","locator":"Data research lab building benchmarks and RL environments for finance and consulting workflows; Series A banner; benchmark dated 2026-08-11; no customers or revenue; no large-scale labelling and multimodal data offering described"}},"rl-environments":{"low":0.638,"central":1.618,"high":4.818,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"Mid-2026 (estimate)","periodEnd":"2026-09-30","source":{"url":"https://www.halluminate.ai/","title":"Halluminate homepage","publisher":"Halluminate","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T5","locator":"Data research lab building benchmarks and RL environments for finance and consulting workflows; Series A banner; benchmark dated 2026-08-11; no customers or revenue"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Method: The page shows no revenue; a directory (RL List, not used as a source) lists a self-reported mid-eight-figure run-rate and a job listing claims about US$42M ARR (both unverified). Central US$35M, range US$15-99M. Builds benchmarks and RL environments for finance and consulting workflows. Main input: Halluminate (T2). Share = company ÷ function total; low and high move only this company's own range and hold the others at central. Halluminate RL environments and agent tasks: US$29.8m (11.6-91.6m) = run-rate US$35m (15-99m) × 85% (the smaller parts of a mixed vendor are varied by ±50% and the largest part absorbs the difference). Function total, annualised US$m (central): 1,839, made of identified candidates 1,739 and an allowance of 100 for vendors outside the register (for example Deccan AI, Shaip, Cogito Tech, Sigma AI, Centaur Labs, Revelo, e2f, Patronus AI, Haize Labs, Lightly, Mostly AI, Abundant, Ethos, Datumo and smaller environment builders). Largest, US$m: mercor 484, surge-ai 414, proximal 140, handshake-ai 136, micro1 130, other candidates 434, vendors outside the register 100. Full table: a local evidence capture. Share 1.618% (range 0.638-4.818%). Source published date: undated (undated)."},"evaluation":{"low":0.193,"central":0.896,"high":3.695,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"Mid-2026 (estimate)","periodEnd":"2026-09-30","source":{"url":"https://www.halluminate.ai/","title":"Halluminate homepage","publisher":"Halluminate","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T5","locator":"Data research lab building benchmarks and RL environments for finance and consulting workflows; Series A banner; benchmark dated 2026-08-11; no customers or revenue"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Method: The page shows no revenue; a directory (RL List, not used as a source) lists a self-reported mid-eight-figure run-rate and a job listing claims about US$42M ARR (both unverified). Central US$35M, range US$15-99M. Builds benchmarks and RL environments for finance and consulting workflows. Main input: Halluminate (T2). Share = company ÷ function total; low and high move only this company's own range and hold the others at central. Halluminate independent model evaluation: US$5.25m (1.12-22.3m) = run-rate US$35m (15-99m) × 15% (the smaller parts of a mixed vendor are varied by ±50% and the largest part absorbs the difference). Function total, annualised US$m (central): 586, made of identified candidates 566 and an allowance of 20 for vendors outside the register (for example Deccan AI, Shaip, Cogito Tech, Sigma AI, Centaur Labs, Revelo, e2f, Patronus AI, Haize Labs, Lightly, Mostly AI, Abundant, Ethos, Datumo and smaller environment builders). Largest, US$m: arena-ai 110, surge-ai 90, mercor 66, scale-ai 45, innodata 33, other candidates 222, vendors outside the register 20. Full table: a local evidence capture. Share 0.896% (range 0.193-3.695%). Source published date: undated (undated)."},"data-tooling":{"participates":false,"evidence":{"url":"https://www.halluminate.ai/","title":"Halluminate homepage","publisher":"Halluminate","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T2","locator":"Data research lab building benchmarks and RL environments for finance and consulting workflows; Series A banner; benchmark dated 2026-08-11; no customers or revenue; no labelling and data software offering described"}}},"screening":{"rl-environments":{"levelLow":1,"levelHigh":2},"evaluation":{"levelLow":1,"levelHigh":2}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"5fb8128cd6e7c5146cde1d41295efd12efba1b157a70c799f57c4e759830cbd9"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"5fb8128cd6e7c5146cde1d41295efd12efba1b157a70c799f57c4e759830cbd9"}}},"gray-swan":{"schema":"5lc-index-inputs/1","key":"models-data-evaluation","entityId":"gray-swan","name":"Gray Swan AI","kind":"private","commercialActivityVerified":false,"shares":{"expert-data":{"participates":false,"evidence":{"url":"https://www.grayswan.ai/","title":"Gray Swan homepage","publisher":"Gray Swan AI","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T2","locator":"Offerings: Shade assessment, Cygnal runtime filtering, Arena red-teaming network, adversarial evaluation for model builders; logos include OpenAI, Anthropic, Google DeepMind (relationship not described); no revenue; no expert data for post-training offering described"}},"labelling-multimodal":{"participates":false,"evidence":{"url":"https://www.grayswan.ai/","title":"Gray Swan homepage","publisher":"Gray Swan AI","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T2","locator":"Offerings: Shade assessment, Cygnal runtime filtering, Arena red-teaming network, adversarial evaluation for model builders; logos include OpenAI, Anthropic, Google DeepMind (relationship not described); no revenue; no large-scale labelling and multimodal data offering described"}},"rl-environments":{"participates":false,"evidence":{"url":"https://www.grayswan.ai/","title":"Gray Swan homepage","publisher":"Gray Swan AI","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T2","locator":"Offerings: Shade assessment, Cygnal runtime filtering, Arena red-teaming network, adversarial evaluation for model builders; logos include OpenAI, Anthropic, Google DeepMind (relationship not described); no revenue; no RL environments and agent tasks offering described"}},"evaluation":{"low":0.527,"central":3.414,"high":9.587,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"Mid-2026 (estimate)","periodEnd":"2026-09-30","source":{"url":"https://www.grayswan.ai/","title":"Gray Swan homepage","publisher":"Gray Swan AI","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T5","locator":"Offerings: Shade assessment, Cygnal runtime filtering, Arena red-teaming network, adversarial evaluation for model builders; logos include OpenAI, Anthropic, Google DeepMind (relationship not described); no revenue"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Method: No revenue is published. Gray Swan sells pre-release adversarial evaluation and a red-teaming network to model builders and a deployment assessment and runtime filter to enterprises; only the model-builder work counts. Central US$20M is an estimate (US$3-60M). Main input: Gray Swan AI (T2). Share = company ÷ function total; low and high move only this company's own range and hold the others at central. Gray Swan AI independent model evaluation: US$20m (3-60m) = run-rate US$20m (3-60m) × 100%. Function total, annualised US$m (central): 586, made of identified candidates 566 and an allowance of 20 for vendors outside the register (for example Deccan AI, Shaip, Cogito Tech, Sigma AI, Centaur Labs, Revelo, e2f, Patronus AI, Haize Labs, Lightly, Mostly AI, Abundant, Ethos, Datumo and smaller environment builders). Largest, US$m: arena-ai 110, surge-ai 90, mercor 66, scale-ai 45, innodata 33, other candidates 222, vendors outside the register 20. Full table: a local evidence capture. Share 3.414% (range 0.527-9.587%). Source published date: undated (undated)."},"data-tooling":{"participates":false,"evidence":{"url":"https://www.grayswan.ai/","title":"Gray Swan homepage","publisher":"Gray Swan AI","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T2","locator":"Offerings: Shade assessment, Cygnal runtime filtering, Arena red-teaming network, adversarial evaluation for model builders; logos include OpenAI, Anthropic, Google DeepMind (relationship not described); no revenue; no labelling and data software offering described"}}},"screening":{"evaluation":{"levelLow":1,"levelHigh":3}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"179e12e532548c08962fdd245bf5395a0900cfbc9e71950ebc85966b27aca590"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"179e12e532548c08962fdd245bf5395a0900cfbc9e71950ebc85966b27aca590"}}},"superannotate":{"schema":"5lc-index-inputs/1","key":"models-data-evaluation","entityId":"superannotate","name":"SuperAnnotate","kind":"private","commercialActivityVerified":true,"shares":{"expert-data":{"low":0.021,"central":0.127,"high":0.443,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"2024-25 evidence, carried to a 2026 estimate","periodEnd":"2025-07-15","source":{"url":"https://www.vcaonline.com/news/2025071524/superannotate-raises-13-5-million-in-new-funding-from-dell-technologies-capital/","title":"SuperAnnotate raises $13.5 million in new funding from Dell Technologies Capital","publisher":"VCA Online","publishedDate":"2025-07-15","accessedDate":"2026-10-22","tier":"T5","locator":"The company says software revenue grew 5x in 2024; hundreds of brands including Databricks, Canva and Motorola Solutions; Series B total $50M; no absolute revenue"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Method: The company says (15 Jul 2025) software revenue grew 5x in 2024, with customers such as Databricks, Canva and Motorola Solutions; no absolute revenue (a third-party estimate of US$45M is not used). Central US$30M is an estimate (US$10-70M). The evidence is over 12 months old, so the engine adds the staleness widening. Main input: VCA Online (T3). Share = company ÷ function total; low and high move only this company's own range and hold the others at central. SuperAnnotate expert data for post-training: US$9m (1.5-31.5m) = run-rate US$30m (10-70m) × 30% (the smaller parts of a mixed vendor are varied by ±50% and the largest part absorbs the difference). Function total, annualised US$m (central): 7,084, made of identified candidates 6,934 and an allowance of 150 for vendors outside the register (for example Deccan AI, Shaip, Cogito Tech, Sigma AI, Centaur Labs, Revelo, e2f, Patronus AI, Haize Labs, Lightly, Mostly AI, Abundant, Ethos, Datumo and smaller environment builders). Largest, US$m: mercor 1,650, surge-ai 1,296, handshake-ai 882, micro1 850, scale-ai 412, other candidates 1,844, vendors outside the register 150. Full table: a local evidence capture. Share 0.127% (range 0.021-0.443%). Source published date: 2025-07-15 (full date)."},"labelling-multimodal":{"participates":false,"evidence":{"url":"https://www.vcaonline.com/news/2025071524/superannotate-raises-13-5-million-in-new-funding-from-dell-technologies-capital/","title":"SuperAnnotate raises $13.5 million in new funding from Dell Technologies Capital","publisher":"VCA Online","publishedDate":"2025-07-15","accessedDate":"2026-10-22","tier":"T3","locator":"The company says software revenue grew 5x in 2024; hundreds of brands including Databricks, Canva and Motorola Solutions; Series B total $50M; no absolute revenue; no large-scale labelling and multimodal data offering described"}},"rl-environments":{"participates":false,"evidence":{"url":"https://www.vcaonline.com/news/2025071524/superannotate-raises-13-5-million-in-new-funding-from-dell-technologies-capital/","title":"SuperAnnotate raises $13.5 million in new funding from Dell Technologies Capital","publisher":"VCA Online","publishedDate":"2025-07-15","accessedDate":"2026-10-22","tier":"T3","locator":"The company says software revenue grew 5x in 2024; hundreds of brands including Databricks, Canva and Motorola Solutions; Series B total $50M; no absolute revenue; no RL environments and agent tasks offering described"}},"evaluation":{"participates":false,"evidence":{"url":"https://www.vcaonline.com/news/2025071524/superannotate-raises-13-5-million-in-new-funding-from-dell-technologies-capital/","title":"SuperAnnotate raises $13.5 million in new funding from Dell Technologies Capital","publisher":"VCA Online","publishedDate":"2025-07-15","accessedDate":"2026-10-22","tier":"T3","locator":"The company says software revenue grew 5x in 2024; hundreds of brands including Databricks, Canva and Motorola Solutions; Series B total $50M; no absolute revenue; no independent model evaluation offering described"}},"data-tooling":{"low":2.216,"central":7.965,"high":19.692,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"2024-25 evidence, carried to a 2026 estimate","periodEnd":"2025-07-15","source":{"url":"https://www.vcaonline.com/news/2025071524/superannotate-raises-13-5-million-in-new-funding-from-dell-technologies-capital/","title":"SuperAnnotate raises $13.5 million in new funding from Dell Technologies Capital","publisher":"VCA Online","publishedDate":"2025-07-15","accessedDate":"2026-10-22","tier":"T5","locator":"The company says software revenue grew 5x in 2024; hundreds of brands including Databricks, Canva and Motorola Solutions; Series B total $50M; no absolute revenue"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Method: The company says (15 Jul 2025) software revenue grew 5x in 2024, with customers such as Databricks, Canva and Motorola Solutions; no absolute revenue (a third-party estimate of US$45M is not used). Central US$30M is an estimate (US$10-70M). The evidence is over 12 months old, so the engine adds the staleness widening. Main input: VCA Online (T3). Share = company ÷ function total; low and high move only this company's own range and hold the others at central. SuperAnnotate labelling and data software: US$21m (5.5-59.5m) = run-rate US$30m (10-70m) × 70% (the smaller parts of a mixed vendor are varied by ±50% and the largest part absorbs the difference). Function total, annualised US$m (central): 264, made of identified candidates 234 and an allowance of 30 for vendors outside the register (for example Deccan AI, Shaip, Cogito Tech, Sigma AI, Centaur Labs, Revelo, e2f, Patronus AI, Haize Labs, Lightly, Mostly AI, Abundant, Ethos, Datumo and smaller environment builders). Largest, US$m: snorkel-ai 37.5, labelbox 37.5, encord 36, roboflow 25, superannotate 21, other candidates 76.7, vendors outside the register 30. Full table: a local evidence capture. Share 7.965% (range 2.216-19.692%). Source published date: 2025-07-15 (full date)."}},"screening":{"expert-data":{"levelLow":1,"levelHigh":2},"data-tooling":{"levelLow":1,"levelHigh":2}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"74e0f40a3802561d3906a0cd01ccb0893a4ee2f77491d089e01fdf56ca63fe79"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"74e0f40a3802561d3906a0cd01ccb0893a4ee2f77491d089e01fdf56ca63fe79"}}},"irregular":{"schema":"5lc-index-inputs/1","key":"models-data-evaluation","entityId":"irregular","name":"Irregular","kind":"private","commercialActivityVerified":false,"shares":{"expert-data":{"participates":false,"evidence":{"url":"https://www.newswire.com/view/content/irregular-raises-80-million-to-set-the-security-standards-for-frontier-ai","title":"Irregular Raises $80 Million to Set the Security Standards for Frontier AI","publisher":"Irregular (press release)","publishedDate":"2025-09-17","accessedDate":"2026-10-22","tier":"T2","locator":"The company says it has reached millions in annual revenue and works with leading AI labs like OpenAI and Anthropic; no expert data for post-training offering described"}},"labelling-multimodal":{"participates":false,"evidence":{"url":"https://www.newswire.com/view/content/irregular-raises-80-million-to-set-the-security-standards-for-frontier-ai","title":"Irregular Raises $80 Million to Set the Security Standards for Frontier AI","publisher":"Irregular (press release)","publishedDate":"2025-09-17","accessedDate":"2026-10-22","tier":"T2","locator":"The company says it has reached millions in annual revenue and works with leading AI labs like OpenAI and Anthropic; no large-scale labelling and multimodal data offering described"}},"rl-environments":{"participates":false,"evidence":{"url":"https://www.newswire.com/view/content/irregular-raises-80-million-to-set-the-security-standards-for-frontier-ai","title":"Irregular Raises $80 Million to Set the Security Standards for Frontier AI","publisher":"Irregular (press release)","publishedDate":"2025-09-17","accessedDate":"2026-10-22","tier":"T2","locator":"The company says it has reached millions in annual revenue and works with leading AI labs like OpenAI and Anthropic; no RL environments and agent tasks offering described"}},"evaluation":{"low":0.52,"central":2.048,"high":6.516,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"Mid-September 2025 evidence, carried to a 2026 estimate","periodEnd":"2025-09-17","source":{"url":"https://www.newswire.com/view/content/irregular-raises-80-million-to-set-the-security-standards-for-frontier-ai","title":"Irregular Raises $80 Million to Set the Security Standards for Frontier AI","publisher":"Irregular (press release)","publishedDate":"2025-09-17","accessedDate":"2026-10-22","tier":"T5","locator":"The company says it has reached millions in annual revenue and works with leading AI labs like OpenAI and Anthropic"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Method: The company says (press release, 17 Sep 2025) it has reached millions in annual revenue and works with labs including OpenAI and Anthropic; it raised US$80M at a reported US$450M valuation. Central US$12M is an estimate (US$3-40M). Evidence is over 12 months old, so the engine adds the staleness widening. Main input: Irregular (press release) (T2). Share = company ÷ function total; low and high move only this company's own range and hold the others at central. Irregular independent model evaluation: US$12m (3-40m) = run-rate US$12m (3-40m) × 100%. Function total, annualised US$m (central): 586, made of identified candidates 566 and an allowance of 20 for vendors outside the register (for example Deccan AI, Shaip, Cogito Tech, Sigma AI, Centaur Labs, Revelo, e2f, Patronus AI, Haize Labs, Lightly, Mostly AI, Abundant, Ethos, Datumo and smaller environment builders). Largest, US$m: arena-ai 110, surge-ai 90, mercor 66, scale-ai 45, innodata 33, other candidates 222, vendors outside the register 20. Full table: a local evidence capture. Share 2.048% (range 0.52-6.516%). Second source(s): TechCrunch 2025-09-17 (T3): Raised $80M, valued at $450M (source close to the deal); its work is cited in security evaluations of OpenAI and Anthropic models; no revenue stated. Source published date: 2025-09-17 (full date)."},"data-tooling":{"participates":false,"evidence":{"url":"https://www.newswire.com/view/content/irregular-raises-80-million-to-set-the-security-standards-for-frontier-ai","title":"Irregular Raises $80 Million to Set the Security Standards for Frontier AI","publisher":"Irregular (press release)","publishedDate":"2025-09-17","accessedDate":"2026-10-22","tier":"T2","locator":"The company says it has reached millions in annual revenue and works with leading AI labs like OpenAI and Anthropic; no labelling and data software offering described"}}},"screening":{"evaluation":{"levelLow":1,"levelHigh":3}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"3811dc44326ef8b073cd3ed79026ff231cca5bb2fe9044156e965c25d94beef8"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"3811dc44326ef8b073cd3ed79026ff231cca5bb2fe9044156e965c25d94beef8"}}},"cloudfactory":{"schema":"5lc-index-inputs/1","key":"models-data-evaluation","entityId":"cloudfactory","name":"CloudFactory","kind":"private","commercialActivityVerified":false,"shares":{"expert-data":{"low":0.004,"central":0.035,"high":0.169,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"Mid-2026 (estimate)","periodEnd":"2026-09-30","source":{"url":"https://www.cloudfactory.com/","title":"CloudFactory homepage","publisher":"CloudFactory","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T5","locator":"Presents an AI platform and consulting firm (data foundation, orchestration, 'Trust & Oversight' with human review); customer logos only (Microsoft, Luminar, Expensify and others) with no stated relationship; no revenue, no headcount"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Method: No revenue is published and no opened source gives a dollar figure, so the whole amount is a T5 judgment. Anchor: the CloudFactory homepage (opened 22 Oct 2026, undated, '(c) 2010-2026') presents an AI platform and consulting firm (data foundation, model and agent orchestration, 'Trust & Oversight' with human review); it does not list data annotation as a service, shows customers as logos only (Microsoft, Luminar, Expensify and others, with no stated relationship) and carries testimonials from Nearmap and LineVision that describe working together without stating payment. Annotation for AI developers is therefore at most a small part of the business. Central US$25M, low US$5M (a few small contracts), high US$80M (a mid-sized annotation business); the 16-fold span is deliberate and the central lies near its geometric midpoint (US$20M). No opened page shows paying customers, revenue or shipments in this sublayer, so commercialActivityVerified is false. Split 10% expert data and 90% labelling: the homepage industries (autonomous vehicles and robotics, damage assessment, insurance) point to image and sensor-data annotation. Main input: none (a judgment; the CloudFactory homepage is qualitative evidence only). Share = company ÷ function total; low and high move only this company's own range and hold the others at central. CloudFactory expert data for post-training: US$2.5m (0.25-12m) = run-rate US$25m (5-80m) × 10% (the smaller parts of a mixed vendor are varied by ±50% and the largest part absorbs the difference). Function total, annualised US$m (central): 7,084, made of identified candidates 6,934 and an allowance of 150 for vendors outside the register (for example Deccan AI, Shaip, Cogito Tech, Sigma AI, Centaur Labs, Revelo, e2f, Patronus AI, Haize Labs, Lightly, Mostly AI, Abundant, Ethos, Datumo and smaller environment builders). Largest, US$m: mercor 1,650, surge-ai 1,296, handshake-ai 882, micro1 850, scale-ai 412, other candidates 1,844, vendors outside the register 150. Full table: a local evidence capture. Share 0.035% (range 0.004-0.169%). Source published date: undated (the page shows no date)."},"labelling-multimodal":{"low":0.244,"central":1.279,"high":4.192,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"Mid-2026 (estimate)","periodEnd":"2026-09-30","source":{"url":"https://www.cloudfactory.com/","title":"CloudFactory homepage","publisher":"CloudFactory","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T5","locator":"Presents an AI platform and consulting firm (data foundation, orchestration, 'Trust & Oversight' with human review); customer logos only (Microsoft, Luminar, Expensify and others) with no stated relationship; no revenue, no headcount"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Method: No revenue is published and no opened source gives a dollar figure, so the whole amount is a T5 judgment. Anchor: the CloudFactory homepage (opened 22 Oct 2026, undated, '(c) 2010-2026') presents an AI platform and consulting firm (data foundation, model and agent orchestration, 'Trust & Oversight' with human review); it does not list data annotation as a service, shows customers as logos only (Microsoft, Luminar, Expensify and others, with no stated relationship) and carries testimonials from Nearmap and LineVision that describe working together without stating payment. Annotation for AI developers is therefore at most a small part of the business. Central US$25M, low US$5M (a few small contracts), high US$80M (a mid-sized annotation business); the 16-fold span is deliberate and the central lies near its geometric midpoint (US$20M). No opened page shows paying customers, revenue or shipments in this sublayer, so commercialActivityVerified is false. Split 10% expert data and 90% labelling: the homepage industries (autonomous vehicles and robotics, damage assessment, insurance) point to image and sensor-data annotation. Main input: none (a judgment; the CloudFactory homepage is qualitative evidence only). Share = company ÷ function total; low and high move only this company's own range and hold the others at central. CloudFactory large-scale labelling and multimodal data: US$22.5m (4.25-76m) = run-rate US$25m (5-80m) × 90% (the smaller parts of a mixed vendor are varied by ±50% and the largest part absorbs the difference). Function total, annualised US$m (central): 1,759, made of identified candidates 1,609 and an allowance of 150 for vendors outside the register (for example Deccan AI, Shaip, Cogito Tech, Sigma AI, Centaur Labs, Revelo, e2f, Patronus AI, Haize Labs, Lightly, Mostly AI, Abundant, Ethos, Datumo and smaller environment builders). Largest, US$m: scale-ai 262, appen 176, taskus 146, rws-trainai 140, telus-digital 124, other candidates 762, vendors outside the register 150. Full table: a local evidence capture. Share 1.279% (range 0.244-4.192%). Source published date: undated (the page shows no date)."},"rl-environments":{"participates":false,"evidence":{"url":"https://www.cloudfactory.com/","title":"CloudFactory homepage","publisher":"CloudFactory","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T2","locator":"Now an AI platform and consulting firm (data foundation, orchestration, human review); logos Microsoft, Luminar, Expensify; no revenue; no RL environments and agent tasks offering described"}},"evaluation":{"participates":false,"evidence":{"url":"https://www.cloudfactory.com/","title":"CloudFactory homepage","publisher":"CloudFactory","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T2","locator":"Now an AI platform and consulting firm (data foundation, orchestration, human review); logos Microsoft, Luminar, Expensify; no revenue; no independent model evaluation offering described"}},"data-tooling":{"participates":false,"evidence":{"url":"https://www.cloudfactory.com/","title":"CloudFactory homepage","publisher":"CloudFactory","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T2","locator":"Now an AI platform and consulting firm (data foundation, orchestration, human review); logos Microsoft, Luminar, Expensify; no revenue; no labelling and data software offering described"}}},"screening":{"expert-data":{"levelLow":1,"levelHigh":2},"labelling-multimodal":{"levelLow":1,"levelHigh":2}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"1b2b68f966dc2383cdd3b7730346e07fe1a0c466f2a98ab5420b92964ebf3c2c"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"1b2b68f966dc2383cdd3b7730346e07fe1a0c466f2a98ab5420b92964ebf3c2c"}}},"afterquery":{"schema":"5lc-index-inputs/1","key":"models-data-evaluation","entityId":"afterquery","name":"AfterQuery","kind":"private","commercialActivityVerified":false,"shares":{"expert-data":{"low":0.018,"central":0.176,"high":1.049,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"Mid-2026 (estimate)","periodEnd":"2026-09-30","source":{"url":"https://www.afterquery.com","title":"AfterQuery homepage","publisher":"AfterQuery","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T5","locator":"Offerings: SFT data, RL with rubrics, agent environments, computer-use trajectories for frontier labs; no named customers or revenue; blog items to 2026-10-15"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Method: No revenue or customer is stated (a directory lists US$30.5M raised). Offers SFT data, RL with rubrics and agent environments for labs. Central US$25M is an estimate (US$5-100M). Main input: AfterQuery (T2). Share = company ÷ function total; low and high move only this company's own range and hold the others at central. AfterQuery expert data for post-training: US$12.5m (1.25-75m) = run-rate US$25m (5-100m) × 50% (the smaller parts of a mixed vendor are varied by ±50% and the largest part absorbs the difference). Function total, annualised US$m (central): 7,084, made of identified candidates 6,934 and an allowance of 150 for vendors outside the register (for example Deccan AI, Shaip, Cogito Tech, Sigma AI, Centaur Labs, Revelo, e2f, Patronus AI, Haize Labs, Lightly, Mostly AI, Abundant, Ethos, Datumo and smaller environment builders). Largest, US$m: mercor 1,650, surge-ai 1,296, handshake-ai 882, micro1 850, scale-ai 412, other candidates 1,844, vendors outside the register 150. Full table: a local evidence capture. Share 0.176% (range 0.018-1.049%). Source published date: undated (undated)."},"labelling-multimodal":{"participates":false,"evidence":{"url":"https://www.afterquery.com","title":"AfterQuery homepage","publisher":"AfterQuery","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T2","locator":"Offerings: SFT data, RL with rubrics, agent environments, computer-use trajectories for frontier labs; no named customers or revenue; blog items to 2026-10-15; no large-scale labelling and multimodal data offering described"}},"rl-environments":{"low":0.068,"central":0.68,"high":3.945,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"Mid-2026 (estimate)","periodEnd":"2026-09-30","source":{"url":"https://www.afterquery.com","title":"AfterQuery homepage","publisher":"AfterQuery","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T5","locator":"Offerings: SFT data, RL with rubrics, agent environments, computer-use trajectories for frontier labs; no named customers or revenue; blog items to 2026-10-15"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Method: No revenue or customer is stated (a directory lists US$30.5M raised). Offers SFT data, RL with rubrics and agent environments for labs. Central US$25M is an estimate (US$5-100M). Main input: AfterQuery (T2). Share = company ÷ function total; low and high move only this company's own range and hold the others at central. AfterQuery RL environments and agent tasks: US$12.5m (1.25-75m) = run-rate US$25m (5-100m) × 50% (the smaller parts of a mixed vendor are varied by ±50% and the largest part absorbs the difference). Function total, annualised US$m (central): 1,839, made of identified candidates 1,739 and an allowance of 100 for vendors outside the register (for example Deccan AI, Shaip, Cogito Tech, Sigma AI, Centaur Labs, Revelo, e2f, Patronus AI, Haize Labs, Lightly, Mostly AI, Abundant, Ethos, Datumo and smaller environment builders). Largest, US$m: mercor 484, surge-ai 414, proximal 140, handshake-ai 136, micro1 130, other candidates 434, vendors outside the register 100. Full table: a local evidence capture. Share 0.68% (range 0.068-3.945%). Source published date: undated (undated)."},"evaluation":{"participates":false,"evidence":{"url":"https://www.afterquery.com","title":"AfterQuery homepage","publisher":"AfterQuery","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T2","locator":"Offerings: SFT data, RL with rubrics, agent environments, computer-use trajectories for frontier labs; no named customers or revenue; blog items to 2026-10-15; no independent model evaluation offering described"}},"data-tooling":{"participates":false,"evidence":{"url":"https://www.afterquery.com","title":"AfterQuery homepage","publisher":"AfterQuery","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T2","locator":"Offerings: SFT data, RL with rubrics, agent environments, computer-use trajectories for frontier labs; no named customers or revenue; blog items to 2026-10-15; no labelling and data software offering described"}}},"screening":{"expert-data":{"levelLow":1,"levelHigh":2},"rl-environments":{"levelLow":1,"levelHigh":2}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"8806f0b0bfec3be514bb2eb72d5937b6846c6134343b1abd7ff5fd3c89bfd57f"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"8806f0b0bfec3be514bb2eb72d5937b6846c6134343b1abd7ff5fd3c89bfd57f"}}},"artificial-analysis":{"schema":"5lc-index-inputs/1","key":"models-data-evaluation","entityId":"artificial-analysis","name":"Artificial Analysis","kind":"private","commercialActivityVerified":false,"shares":{"expert-data":{"participates":false,"evidence":{"url":"https://artificialanalysis.ai/","title":"Artificial Analysis homepage","publisher":"Artificial Analysis","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T2","locator":"Model and provider benchmarks, Optima custom benchmarks; changelog to 22 October 2026; how it earns revenue is not stated; no expert data for post-training offering described"}},"labelling-multimodal":{"participates":false,"evidence":{"url":"https://artificialanalysis.ai/","title":"Artificial Analysis homepage","publisher":"Artificial Analysis","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T2","locator":"Model and provider benchmarks, Optima custom benchmarks; changelog to 22 October 2026; how it earns revenue is not stated; no large-scale labelling and multimodal data offering described"}},"rl-environments":{"participates":false,"evidence":{"url":"https://artificialanalysis.ai/","title":"Artificial Analysis homepage","publisher":"Artificial Analysis","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T2","locator":"Model and provider benchmarks, Optima custom benchmarks; changelog to 22 October 2026; how it earns revenue is not stated; no RL environments and agent tasks offering described"}},"evaluation":{"low":0.523,"central":2.56,"high":8.053,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"Mid-2026 (estimate)","periodEnd":"2026-09-30","source":{"url":"https://artificialanalysis.ai/","title":"Artificial Analysis homepage","publisher":"Artificial Analysis","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T5","locator":"Model and provider benchmarks, Optima custom benchmarks; changelog to 22 October 2026; how it earns revenue is not stated"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Method: How Artificial Analysis earns revenue is not stated on its pages (custom benchmarks are listed). Central US$15M is an estimate (US$3-50M). Main input: Artificial Analysis (T2). Share = company ÷ function total; low and high move only this company's own range and hold the others at central. Artificial Analysis independent model evaluation: US$15m (3-50m) = run-rate US$15m (3-50m) × 100%. Function total, annualised US$m (central): 586, made of identified candidates 566 and an allowance of 20 for vendors outside the register (for example Deccan AI, Shaip, Cogito Tech, Sigma AI, Centaur Labs, Revelo, e2f, Patronus AI, Haize Labs, Lightly, Mostly AI, Abundant, Ethos, Datumo and smaller environment builders). Largest, US$m: arena-ai 110, surge-ai 90, mercor 66, scale-ai 45, innodata 33, other candidates 222, vendors outside the register 20. Full table: a local evidence capture. Share 2.56% (range 0.523-8.053%). Source published date: undated (undated)."},"data-tooling":{"participates":false,"evidence":{"url":"https://artificialanalysis.ai/","title":"Artificial Analysis homepage","publisher":"Artificial Analysis","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T2","locator":"Model and provider benchmarks, Optima custom benchmarks; changelog to 22 October 2026; how it earns revenue is not stated; no labelling and data software offering described"}}},"screening":{"evaluation":{"levelLow":1,"levelHigh":3}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"0f2224a2bae224789b4e74221874b3c69ed7fb8d974fc9bd7bdda81a9d58e56e"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"0f2224a2bae224789b4e74221874b3c69ed7fb8d974fc9bd7bdda81a9d58e56e"}}},"datacurve":{"schema":"5lc-index-inputs/1","key":"models-data-evaluation","entityId":"datacurve","name":"Datacurve","kind":"private","commercialActivityVerified":false,"shares":{"expert-data":{"low":0.01,"central":0.127,"high":0.757,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"Mid-2026 (estimate)","periodEnd":"2026-09-30","source":{"url":"https://datacurve.ai/","title":"Datacurve homepage","publisher":"Datacurve","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T5","locator":"Data collection research lab; DeepSWE coding benchmark; no named customers or revenue"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Method: No revenue or customer is stated (a directory lists US$17.7M raised). Coding data and environments. Central US$20M is an estimate (US$3-80M). Main input: Datacurve (T2). Share = company ÷ function total; low and high move only this company's own range and hold the others at central. Datacurve expert data for post-training: US$9m (0.68-54m) = run-rate US$20m (3-80m) × 45% (the smaller parts of a mixed vendor are varied by ±50% and the largest part absorbs the difference). Function total, annualised US$m (central): 7,084, made of identified candidates 6,934 and an allowance of 150 for vendors outside the register (for example Deccan AI, Shaip, Cogito Tech, Sigma AI, Centaur Labs, Revelo, e2f, Patronus AI, Haize Labs, Lightly, Mostly AI, Abundant, Ethos, Datumo and smaller environment builders). Largest, US$m: mercor 1,650, surge-ai 1,296, handshake-ai 882, micro1 850, scale-ai 412, other candidates 1,844, vendors outside the register 150. Full table: a local evidence capture. Share 0.127% (range 0.01-0.757%). Source published date: undated (undated)."},"labelling-multimodal":{"participates":false,"evidence":{"url":"https://datacurve.ai/","title":"Datacurve homepage","publisher":"Datacurve","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T2","locator":"Data collection research lab; DeepSWE coding benchmark; no named customers or revenue; no large-scale labelling and multimodal data offering described"}},"rl-environments":{"low":0.053,"central":0.598,"high":3.281,"unit":"percent","basis":"Share of the function's annualised gross revenue run-rate in US dollars, before payments to contributors (latest month × 12 or latest quarter × 4)","period":"Mid-2026 (estimate)","periodEnd":"2026-09-30","source":{"url":"https://datacurve.ai/","title":"Datacurve homepage","publisher":"Datacurve","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T5","locator":"Data collection research lab; DeepSWE coding benchmark; no named customers or revenue"},"note":"5LayerCake estimate (T5), full-inputs pass of 22 October 2026. Method: No revenue or customer is stated (a directory lists US$17.7M raised). Coding data and environments. Central US$20M is an estimate (US$3-80M). Main input: Datacurve (T2). Share = company ÷ function total; low and high move only this company's own range and hold the others at central. Datacurve RL environments and agent tasks: US$11m (0.97-62m) = run-rate US$20m (3-80m) × 55% (the smaller parts of a mixed vendor are varied by ±50% and the largest part absorbs the difference). Function total, annualised US$m (central): 1,839, made of identified candidates 1,739 and an allowance of 100 for vendors outside the register (for example Deccan AI, Shaip, Cogito Tech, Sigma AI, Centaur Labs, Revelo, e2f, Patronus AI, Haize Labs, Lightly, Mostly AI, Abundant, Ethos, Datumo and smaller environment builders). Largest, US$m: mercor 484, surge-ai 414, proximal 140, handshake-ai 136, micro1 130, other candidates 434, vendors outside the register 100. Full table: a local evidence capture. Share 0.598% (range 0.053-3.281%). Source published date: undated (undated)."},"evaluation":{"participates":false,"evidence":{"url":"https://datacurve.ai/","title":"Datacurve homepage","publisher":"Datacurve","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T2","locator":"Data collection research lab; DeepSWE coding benchmark; no named customers or revenue; no independent model evaluation offering described"}},"data-tooling":{"participates":false,"evidence":{"url":"https://datacurve.ai/","title":"Datacurve homepage","publisher":"Datacurve","publishedDate":"undated","accessedDate":"2026-10-22","tier":"T2","locator":"Data collection research lab; DeepSWE coding benchmark; no named customers or revenue; no labelling and data software offering described"}}},"screening":{"expert-data":{"levelLow":1,"levelHigh":2},"rl-environments":{"levelLow":1,"levelHigh":2}},"checks":{"sourceSupport":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"277afd5f9a0a209f9465858f95a833675b1620edd81a8725372f6a1b9dcb8abf"},"consistency":{"by":"Claude check pass, Sonnet 5.5","date":"2026-10-22","passed":true,"sha256":"277afd5f9a0a209f9465858f95a833675b1620edd81a8725372f6a1b9dcb8abf"}}}},"ratings":{"mercor":{"a":{"schema":"5lc-index-rating/1","key":"models-data-evaluation","entityId":"mercor","rater":"a","model":"claude-opus-5-5","ratedAt":"2026-10-22","functions":{"expert-data":{"alternatives":{"level":2,"evidence":[{"url":"https://techcrunch.com/2026/09/22/snorkel-ai-triples-valuation-to-3-5b-as-demand-for-ai-training-data-booms/","date":"2026-09-22","support":"TechCrunch names Mercor (US$2bn gross), Handshake (US$1bn) and micro1 (US$500m) as scaled peers, and Snorkel at US$375m run-rate (company says) (paragraphs 4-5); several large alternatives to Mercor. Full date."},{"url":"https://www.pymnts.com/artificial-intelligence-2/2025/surge-ai-reportedly-targeting-1-billion-in-first-ever-fundraise/","date":"2025-07-01","support":"Citing Reuters: Surge took in more than US$1bn of 2024 revenue, above Scale's US$870m (paragraph 4); with Handshake and micro1, about four to five vendors operate at frontier expert-data scale. Full date."}]},"switchingTime":{"level":2,"evidence":[{"url":"https://techcrunch.com/2026/08/20/ai-data-startup-micro1-reaches-500m-gross-run-rate-amid-ai-training-boom/","date":"2026-08-20","support":"micro1 grew its gross run-rate from US$100m to US$500m in eight months (paragraph 1): rivals of Mercor can add expert-data capacity within months, though not weeks. Full date."},{"url":"https://techcrunch.com/2025/10/27/mercor-quintuples-valuation-to-10b-with-350m-series-c/","date":"2025-10-27","support":"Mercor says it has more than 30,000 experts and pays over US$1.5m a day (paragraph 9); re-recruiting that pool elsewhere would take months. Full date."}]},"capabilityGap":{"level":2,"evidence":[{"url":"https://newsletter.semianalysis.com/p/rl-environments-and-rl-for-science","date":"2026-01-06","support":"SemiAnalysis names Surge the largest data contractor, revenue 'closer to $1B ARR' (vendor section); the best alternative is of similar order but somewhat smaller than Mercor's US$2bn gross. Full date."},{"url":"https://techcrunch.com/2026/07/08/these-ai-startups-are-growing-revenue-at-faster-and-faster-rates/","date":"2026-07-08","support":"Mercor's CEO says it went from US$1bn to US$2bn gross annualised revenue in about four months to June 2026 (paragraph 5): the market re-staffs expert work for Mercor-sized volumes within months. Full date."}]}},"rl-environments":{"alternatives":{"level":2,"evidence":[{"url":"https://techcrunch.com/2025/09/16/silicon-valley-bets-big-on-environments-to-train-ai-agents/","date":"2025-09-16","support":"Surge built an internal environments team and Mercor pitches domain environments (paragraphs 15, 17); with Handshake, Turing and Snorkel, a handful of vendors match Mercor's breadth, though 35+ small builders exist. Full date."},{"url":"https://newsletter.semianalysis.com/p/rl-environments-and-rl-for-science","date":"2026-01-06","support":"SemiAnalysis counts \"more than 35 companies\" building RL environments, mostly seed-stage, each serving 1-3 customers (vendor section); Mercor faces many substitutes. Full date."}]},"switchingTime":{"level":2,"evidence":[{"url":"https://newsletter.semianalysis.com/p/rl-environments-and-rl-for-science","date":"2026-01-06","support":"SemiAnalysis says most environment vendors sell under exclusive contracts and UI-gym sites cost about US$20,000 each as one-time purchases (vendor and pricing sections); replacing Mercor's pipeline means commissioning new builds. Full date."},{"url":"https://techcrunch.com/2025/09/16/silicon-valley-bets-big-on-environments-to-train-ai-agents/","date":"2025-09-16","support":"Ross Taylor warns \"people are underestimating how difficult it is to scale environments\" (paragraph 32); a lab replacing Mercor's environment output would need months to re-commission it. Full date."}]},"capabilityGap":{"level":2,"evidence":[{"url":"https://techcrunch.com/2025/09/16/silicon-valley-bets-big-on-environments-to-train-ai-agents/","date":"2025-09-16","support":"Surge's CEO reports a significant increase in environment demand and an internal build team (paragraph 15): the closest alternative has similar reach but less disclosed scale than Mercor. Full date."},{"url":"https://epoch.ai/gradient-updates/state-of-rl-envs","date":"2026-01-12","support":"Epoch's FAQ says the data providers' main value is that \"they have the guys\" and that maintaining quality while scaling is the top bottleneck (sections 'Which companies build RL Environments?', 'top priorities'); relevant to Mercor. Full date."}]}},"evaluation":{"alternatives":{"level":1,"evidence":[{"url":"https://techcrunch.com/2026/06/29/arena-the-ai-leaderboard-everyone-uses-is-now-a-100m-business/","date":"2026-06-29","support":"TechCrunch says human-labelling firms Mercor, Surge and Scale AI compete with Arena \"for the same dollar\" in evaluation (paragraph 7); Mercor's evaluation work has several qualified substitutes. Full date."},{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02-26","support":"Innodata's FY2025 10-K (Item 1, Competition) names Appen, CloudFactory, Surge AI, Invisible, Turing, Mercor, TELUS Digital and Scale AI as AI-data competitors; Mercor work has many qualified homes. Date: EDGAR filing date."},{"url":"https://news.bgov.com/legal-ops-and-tech/scale-unveils-new-ai-model-leaderboard-system-to-rival-lmarena","date":"2025-09-22","support":"Scale runs public evaluation leaderboards with contributors in over 100 countries (paragraphs 1-2); another qualified evaluation supplier besides Surge and Arena. Full date."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02-26","support":"Innodata's 10-K (Item 1A) says customer agreements are generally terminable on about 30-90 days' notice, including on selecting another provider; Mercor volume can be moved on that timescale. Date: EDGAR filing date."},{"url":"https://fortune.com/2025/06/19/openai-is-phasing-out-scale-ai-work-following-startups-meta-deal/","date":"2025-06-19","support":"OpenAI phased out Scale AI work within days of the Meta deal and relied increasingly on other providers such as Mercor (paragraphs 1, 9); Mercor's expert-data work is similarly portable. Full date."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://techcrunch.com/2026/06/29/arena-the-ai-leaderboard-everyone-uses-is-now-a-100m-business/","date":"2026-06-29","support":"TechCrunch says human-labelling firms Mercor, Surge and Scale AI compete with Arena \"for the same dollar\" in evaluation (paragraph 7); Mercor's evaluation work has several qualified substitutes. Full date."}]}}}},"b":{"schema":"5lc-index-rating/1","key":"models-data-evaluation","entityId":"mercor","rater":"b","model":"claude-sonnet-5-5","ratedAt":"2026-10-22","functions":{"expert-data":{"alternatives":{"level":2,"evidence":[{"url":"https://techcrunch.com/2026/09/22/snorkel-ai-triples-valuation-to-3-5b-as-demand-for-ai-training-data-booms/","date":"2026-09-22","support":"Lists Mercor (about $2bn gross annualised), Handshake (about $1bn) and micro1 (about $500M gross) as similar AI-data companies; experts take roughly 60-70% of their top line."},{"url":"https://newsletter.semianalysis.com/p/rl-environments-and-rl-for-science","date":"2026-01-06","support":"Names Mercor, Handshake, Surge and Aboda as expert-hiring firms and puts Surge 'closer to $1B ARR' (authors' estimate); counting Surge, Handshake, micro1 and Scale gives about four peers at comparable scale (my count, 6 Jan 2026)."}]},"switchingTime":{"level":2,"evidence":[{"url":"https://techcrunch.com/2026/07/08/these-ai-startups-are-growing-revenue-at-faster-and-faster-rates/","date":"2026-07-08","support":"Mercor itself went from $500M (Sept 2025) to $1bn to over $2bn gross annualised (June 2026, company says), so peers have shown they can add $0.5-1bn within about a year."},{"url":"https://techcrunch.com/2026/04/09/after-data-breach-10b-valued-startup-mercor-is-having-a-month","date":"2026-04-09","support":"Meta paused its Mercor contracts indefinitely after the March 2026 breach while OpenAI kept working with Mercor, so labs can re-route volume but at lab-by-lab pace (9 Apr 2026)."}]},"capabilityGap":{"level":2,"evidence":[{"url":"https://newsletter.semianalysis.com/p/rl-environments-and-rl-for-science","date":"2026-01-06","support":"Surge is described as the larger, more established player, with no ranking of quality among Mercor, Handshake and Surge (6 Jan 2026); the nearest peers are reported at roughly half (Handshake, micro1) to nine-tenths (Surge, 2025 estimate) of Mercor's gross run-rate."},{"url":"https://techcrunch.com/2026/09/22/snorkel-ai-triples-valuation-to-3-5b-as-demand-for-ai-training-data-booms/","date":"2026-09-22","support":"Lists Mercor (about $2bn gross annualised), Handshake (about $1bn) and micro1 (about $500M gross) as similar AI-data companies; experts take roughly 60-70% of their top line."}]}},"rl-environments":{"alternatives":{"level":1,"evidence":[{"url":"https://newsletter.semianalysis.com/p/rl-environments-and-rl-for-science","date":"2026-01-06","support":"More than 35 companies sell RL environments (mostly seed-stage) and Anthropic likely works with more than a dozen of them (SemiAnalysis, 6 Jan 2026)."},{"url":"https://epoch.ai/gradient-updates/state-of-rl-envs","date":"2026-01-12","support":"Epoch lists specialist startups, the data vendors Mercor, Surge, Handshake and Turing, in-house lab teams and product-company partners as sources of environments (12 Jan 2026)."}]},"switchingTime":{"level":2,"evidence":[{"url":"https://epoch.ai/gradient-updates/state-of-rl-envs","date":"2026-01-12","support":"Contracts are often six to seven figures per quarter and exclusive deals cost roughly 4-5x non-exclusive ones; Mercor reports the largest gross run-rate among data vendors (over $2bn, company says), so replacing its environment volume means several vendors ramping at once (12 Jan 2026; my inference)."},{"url":"https://newsletter.semianalysis.com/p/rl-environments-and-rl-for-science","date":"2026-01-06","support":"Almost all environment companies are seed-stage with fewer than 20 staff, so volume moves mainly to the other large data vendors (6 Jan 2026)."}]},"capabilityGap":{"level":2,"evidence":[{"url":"https://epoch.ai/gradient-updates/state-of-rl-envs","date":"2026-01-12","support":"'Maintaining quality while scaling is the number one bottleneck' for environment vendors, so only the other large data foundries can match Mercor's volume (12 Jan 2026)."},{"url":"https://newsletter.semianalysis.com/p/rl-environments-and-rl-for-science","date":"2026-01-06","support":"Almost all environment startups are seed-stage with under 20 staff; Epoch names Surge, Handshake and Turing as the other data vendors selling environments (6 Jan 2026)."}]}},"evaluation":{"alternatives":{"level":1,"evidence":[{"url":"https://labs.scale.com/leaderboard","date":"undated","support":"Scale Labs publishes 19 active leaderboards (page undated; the footer shows only a 2026 copyright)."},{"url":"https://www.prolific.com/","date":"undated","support":"Prolific sells human feedback for preference tuning, safety evals and benchmarks and lists a HUMAINE leaderboard (page undated; the footer shows only a 2026 copyright)."},{"url":"https://nebius.com/newsroom/nebius-welcomes-bezos-expeditions-as-lead-investor-in-ai-data-business-toloka","date":"2025-05-07","support":"Toloka offers agent safety evaluation and red teaming to labs (Nebius announcement, 7 May 2025)."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://mercor.com/blog/introducing-apex-ai-productivity-index/","date":"2025-10-01","support":"Mercor built its APEX v1 test set with about 100 experts in four professions (200 of 300 prompts approved), suggesting a comparable expert-written set is a matter of months (my inference, 1 Oct 2025)."},{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02","support":"Innodata says its customer agreements are generally terminable on 30 to 90 days' notice and do not commit customers to future purchases (risk factors). Date: month only."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://scale.com/blog/scales-next-era-building-for-2026","date":"2026-01-22","support":"Scale's lab says its lab introduced 15 new benchmarks and published over 450 evaluations of more than 50 models, a scale other vendors also reach (company blog, 22 Jan 2026)."},{"url":"https://welodata.ai/","date":"undated","support":"Welo Data lists evaluations, benchmarking, model selection and red teaming among its services (page undated; the footer shows only a 2026 copyright)."}]}}}}},"surge-ai":{"a":{"schema":"5lc-index-rating/1","key":"models-data-evaluation","entityId":"surge-ai","rater":"a","model":"claude-opus-5-5","ratedAt":"2026-10-22","functions":{"expert-data":{"alternatives":{"level":2,"evidence":[{"url":"https://techcrunch.com/2026/09/22/snorkel-ai-triples-valuation-to-3-5b-as-demand-for-ai-training-data-booms/","date":"2026-09-22","support":"TechCrunch names Mercor (US$2bn gross), Handshake (US$1bn) and micro1 (US$500m) as scaled peers, and Snorkel at US$375m run-rate (company says) (paragraphs 4-5); several large alternatives to Surge. Full date."},{"url":"https://techcrunch.com/2025/12/04/micro1-a-scale-ai-competitor-touts-crossing-100m-arr/","date":"2025-12-04","support":"micro1 describes Mercor and Surge as the rivals it has not yet matched (paragraphs 4-5): only a few vendors are at Surge's tier. Full date."}]},"switchingTime":{"level":2,"evidence":[{"url":"https://techcrunch.com/2026/08/20/ai-data-startup-micro1-reaches-500m-gross-run-rate-amid-ai-training-boom/","date":"2026-08-20","support":"micro1 grew its gross run-rate from US$100m to US$500m in eight months (paragraph 1): rivals of Surge can add expert-data capacity within months, though not weeks. Full date."},{"url":"https://techcrunch.com/2026/07/08/these-ai-startups-are-growing-revenue-at-faster-and-faster-rates/","date":"2026-07-08","support":"Mercor's CEO says it went from US$1bn to US$2bn gross annualised revenue in about four months to June 2026 (paragraph 5): the market re-staffs expert work for Surge-sized volumes within months. Full date."}]},"capabilityGap":{"level":2,"evidence":[{"url":"https://www.pymnts.com/artificial-intelligence-2/2025/surge-ai-reportedly-targeting-1-billion-in-first-ever-fundraise/","date":"2025-07-01","support":"Citing Reuters: Surge's 2024 revenue exceeded US$1bn, larger than Scale's (paragraph 4); Mercor has since passed it in gross run-rate, so the best alternative is close in scale. Full date."},{"url":"https://techcrunch.com/2026/07/08/these-ai-startups-are-growing-revenue-at-faster-and-faster-rates/","date":"2026-07-08","support":"Mercor's CEO says it went from US$1bn to US$2bn gross annualised revenue in about four months to June 2026 (paragraph 5): the market re-staffs expert work for Surge-sized volumes within months. Full date."}]}},"rl-environments":{"alternatives":{"level":2,"evidence":[{"url":"https://techcrunch.com/2025/09/16/silicon-valley-bets-big-on-environments-to-train-ai-agents/","date":"2025-09-16","support":"Mercor pitches domain environments for coding, health and law, and Scale is adapting to build them (paragraphs 17, 19); with Handshake and Turing a few vendors match Surge's reach, beside 35+ small builders. Full date."},{"url":"https://newsletter.semianalysis.com/p/rl-environments-and-rl-for-science","date":"2026-01-06","support":"SemiAnalysis counts \"more than 35 companies\" building RL environments, mostly seed-stage, each serving 1-3 customers (vendor section); Surge faces many substitutes. Full date."}]},"switchingTime":{"level":2,"evidence":[{"url":"https://newsletter.semianalysis.com/p/rl-environments-and-rl-for-science","date":"2026-01-06","support":"SemiAnalysis says most environment vendors sell under exclusive contracts and UI-gym sites cost about US$20,000 each as one-time purchases (vendor and pricing sections); replacing Surge's pipeline means commissioning new builds. Full date."},{"url":"https://techcrunch.com/2025/09/16/silicon-valley-bets-big-on-environments-to-train-ai-agents/","date":"2025-09-16","support":"Ross Taylor warns \"people are underestimating how difficult it is to scale environments\" (paragraph 32); a lab replacing Surge's environment output would need months to re-commission it. Full date."}]},"capabilityGap":{"level":2,"evidence":[{"url":"https://techcrunch.com/2026/07/08/these-ai-startups-are-growing-revenue-at-faster-and-faster-rates/","date":"2026-07-08","support":"Mercor's CEO reports US$2bn gross run-rate by June 2026 (paragraph 5): the best alternative environment supplier is of comparable scale, a small gap. Full date."},{"url":"https://epoch.ai/gradient-updates/state-of-rl-envs","date":"2026-01-12","support":"Epoch's FAQ says the data providers' main value is that \"they have the guys\" and that maintaining quality while scaling is the top bottleneck (sections 'Which companies build RL Environments?', 'top priorities'); relevant to Surge. Full date."}]}},"evaluation":{"alternatives":{"level":1,"evidence":[{"url":"https://techcrunch.com/2026/06/29/arena-the-ai-leaderboard-everyone-uses-is-now-a-100m-business/","date":"2026-06-29","support":"TechCrunch says human-labelling firms Mercor, Surge and Scale AI compete with Arena \"for the same dollar\" in evaluation (paragraph 7); Surge's evaluation work has several qualified substitutes. Full date."},{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02-26","support":"Innodata's FY2025 10-K (Item 1, Competition) names Appen, CloudFactory, Surge AI, Invisible, Turing, Mercor, TELUS Digital and Scale AI as AI-data competitors; Surge work has many qualified homes. Date: EDGAR filing date."},{"url":"https://www.prolific.com/resources/prolific-in-2025-more-precision-larger-scale-and-stronger-safeguards-for-quality-human-data","date":"2025-12-16","support":"Prolific says nearly 27,000 evaluators compared 29 frontier models in its HUMAINE study (paragraph 7); human-evaluation capacity exists outside Surge. Full date."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02-26","support":"Innodata's 10-K (Item 1A) says customer agreements are generally terminable on about 30-90 days' notice, including on selecting another provider; Surge volume can be moved on that timescale. Date: EDGAR filing date."},{"url":"https://fortune.com/2025/06/19/openai-is-phasing-out-scale-ai-work-following-startups-meta-deal/","date":"2025-06-19","support":"OpenAI phased out Scale AI work within days of the Meta deal and relied increasingly on other providers such as Mercor (paragraphs 1, 9); Surge's expert-data work is similarly portable. Full date."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://techcrunch.com/2026/06/29/arena-the-ai-leaderboard-everyone-uses-is-now-a-100m-business/","date":"2026-06-29","support":"TechCrunch says human-labelling firms Mercor, Surge and Scale AI compete with Arena \"for the same dollar\" in evaluation (paragraph 7); Surge's evaluation work has several qualified substitutes. Full date."}]}}}},"b":{"schema":"5lc-index-rating/1","key":"models-data-evaluation","entityId":"surge-ai","rater":"b","model":"claude-sonnet-5-5","ratedAt":"2026-10-22","functions":{"expert-data":{"alternatives":{"level":2,"evidence":[{"url":"https://techcrunch.com/2026/09/22/snorkel-ai-triples-valuation-to-3-5b-as-demand-for-ai-training-data-booms/","date":"2026-09-22","support":"Lists Mercor (about $2bn gross annualised), Handshake (about $1bn) and micro1 (about $500M gross) as similar AI-data companies; experts take roughly 60-70% of their top line."},{"url":"https://siliconangle.com/2025/07/01/data-labeling-startup-surge-ai-seeking-1b-first-capital-raise-reports-say/","date":"2025-07-01","support":"Surge's reported rival Scale sold $870M in 2024 and lost OpenAI and Google to Surge; Anthropic also uses Surge (1 Jul 2025, reported)."}]},"switchingTime":{"level":2,"evidence":[{"url":"https://www.storyboard18.com/how-it-works/google-to-sever-ties-with-scale-ai-following-metas-significant-stake-70359.htm","date":"2025-06-16","support":"Google reportedly planned to end about $200M a year of Scale spend after Meta's stake; Handshake's CEO said demand 'tripled overnight' and Labelbox expected hundreds of millions in new revenue (report of 16 Jun 2025)."},{"url":"https://siliconangle.com/2025/07/01/data-labeling-startup-surge-ai-seeking-1b-first-capital-raise-reports-say/","date":"2025-07-01","support":"Surge reportedly won former Scale customers including OpenAI and Google by 1 July 2025, under three weeks after Meta's stake was confirmed on 13 June (SiliconANGLE, citing reports)."},{"url":"https://pureai.com/articles/2025/06/20/meta-invests-in-scale-ai-while-openai-and-google-step-back.aspx","date":"2025-06-20","support":"OpenAI's move away from Scale was gradual, over roughly the prior year, which suggests several months for a large lab to complete a switch (20 Jun 2025)."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://newsletter.semianalysis.com/p/rl-environments-and-rl-for-science","date":"2026-01-06","support":"Mercor and Handshake hire experts at comparable scale and no source ranks Surge's quality above them; SemiAnalysis calls Surge 'the more established and larger player' on the authors' own estimate (6 Jan 2026)."},{"url":"https://techcrunch.com/2026/07/08/these-ai-startups-are-growing-revenue-at-faster-and-faster-rates/","date":"2026-07-08","support":"Mercor now reports over $2bn gross annualised (June 2026, company says), larger than Surge's last reported level, so the best alternative has no capacity shortfall."}]}},"rl-environments":{"alternatives":{"level":1,"evidence":[{"url":"https://newsletter.semianalysis.com/p/rl-environments-and-rl-for-science","date":"2026-01-06","support":"More than 35 companies sell RL environments (mostly seed-stage) and Anthropic likely works with more than a dozen of them (SemiAnalysis, 6 Jan 2026)."},{"url":"https://epoch.ai/gradient-updates/state-of-rl-envs","date":"2026-01-12","support":"Epoch lists specialist startups, the data vendors Mercor, Surge, Handshake and Turing, in-house lab teams and product-company partners as sources of environments (12 Jan 2026)."}]},"switchingTime":{"level":2,"evidence":[{"url":"https://newsletter.semianalysis.com/p/rl-environments-and-rl-for-science","date":"2026-01-06","support":"OpenAI is building an in-house human-data team to reduce reliance on Surge, Mercor and Handshake, while Anthropic buys from many vendors; replacing Surge's volume means moving several lab contracts (6 Jan 2026)."},{"url":"https://epoch.ai/gradient-updates/state-of-rl-envs","date":"2026-01-12","support":"Contracts of six to seven figures per quarter allow re-sourcing each quarter, but quality at scale is the bottleneck (12 Jan 2026)."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://epoch.ai/gradient-updates/state-of-rl-envs","date":"2026-01-12","support":"Quality at scale is the main bottleneck for all vendors; interviewees rank robustness to reward hacking first and say domain knowledge matters more than ML skill (12 Jan 2026). No vendor is described as uniquely capable."},{"url":"https://newsletter.semianalysis.com/p/rl-environments-and-rl-for-science","date":"2026-01-06","support":"Reports 'significant diversity in quality or focus' among vendors, with Turing on UI gyms and others on software engineering, rather than a single leader (6 Jan 2026)."}]}},"evaluation":{"alternatives":{"level":1,"evidence":[{"url":"https://labs.scale.com/leaderboard","date":"undated","support":"Scale Labs publishes 19 active leaderboards (page undated; the footer shows only a 2026 copyright)."},{"url":"https://www.prolific.com/","date":"undated","support":"Prolific sells human feedback for preference tuning, safety evals and benchmarks and lists a HUMAINE leaderboard (page undated; the footer shows only a 2026 copyright)."},{"url":"https://nebius.com/newsroom/nebius-welcomes-bezos-expeditions-as-lead-investor-in-ai-data-business-toloka","date":"2025-05-07","support":"Toloka offers agent safety evaluation and red teaming to labs (Nebius announcement, 7 May 2025)."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://mercor.com/blog/introducing-apex-ai-productivity-index/","date":"2025-10-01","support":"Mercor built its APEX v1 test set with about 100 experts in four professions (200 of 300 prompts approved), suggesting a comparable expert-written set is a matter of months (my inference, 1 Oct 2025)."},{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02","support":"Innodata says its customer agreements are generally terminable on 30 to 90 days' notice and do not commit customers to future purchases (risk factors). Date: month only."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://scale.com/blog/scales-next-era-building-for-2026","date":"2026-01-22","support":"Scale's lab says its lab introduced 15 new benchmarks and published over 450 evaluations of more than 50 models, a scale other vendors also reach (company blog, 22 Jan 2026)."},{"url":"https://welodata.ai/","date":"undated","support":"Welo Data lists evaluations, benchmarking, model selection and red teaming among its services (page undated; the footer shows only a 2026 copyright)."}]}}}}},"handshake-ai":{"a":{"schema":"5lc-index-rating/1","key":"models-data-evaluation","entityId":"handshake-ai","rater":"a","model":"claude-opus-5-5","ratedAt":"2026-10-22","functions":{"expert-data":{"alternatives":{"level":1,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02-26","support":"Innodata's FY2025 10-K (Item 1, Competition) names Appen, CloudFactory, Surge AI, Invisible, Turing, Mercor, TELUS Digital and Scale AI as AI-data competitors; Handshake work has many qualified homes. Date: EDGAR filing date."},{"url":"https://techcrunch.com/2026/09/22/snorkel-ai-triples-valuation-to-3-5b-as-demand-for-ai-training-data-booms/","date":"2026-09-22","support":"TechCrunch names Mercor (US$2bn gross), Handshake (US$1bn) and micro1 (US$500m) as scaled peers, and Snorkel at US$375m run-rate (company says) (paragraphs 4-5); several large alternatives to Handshake. Full date."}]},"switchingTime":{"level":2,"evidence":[{"url":"https://techcrunch.com/2026/08/20/ai-data-startup-micro1-reaches-500m-gross-run-rate-amid-ai-training-boom/","date":"2026-08-20","support":"micro1 grew its gross run-rate from US$100m to US$500m in eight months (paragraph 1): rivals of Handshake can add expert-data capacity within months, though not weeks. Full date."},{"url":"https://techcrunch.com/2026/07/08/these-ai-startups-are-growing-revenue-at-faster-and-faster-rates/","date":"2026-07-08","support":"Mercor's CEO says it went from US$1bn to US$2bn gross annualised revenue in about four months to June 2026 (paragraph 5): the market re-staffs expert work for Handshake-sized volumes within months. Full date."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://techcrunch.com/2026/08/20/ai-data-startup-micro1-reaches-500m-gross-run-rate-amid-ai-training-boom/","date":"2026-08-20","support":"Mercor reached US$2bn gross annualised revenue versus Handshake's US$1bn (paragraphs 7-8): a larger alternative exists. Full date."}]}},"rl-environments":{"alternatives":{"level":1,"evidence":[{"url":"https://newsletter.semianalysis.com/p/rl-environments-and-rl-for-science","date":"2026-01-06","support":"SemiAnalysis counts \"more than 35 companies\" building RL environments, mostly seed-stage, each serving 1-3 customers (vendor section); Handshake faces many substitutes. Full date."},{"url":"https://epoch.ai/gradient-updates/state-of-rl-envs","date":"2026-01-12","support":"Epoch's RL-environment FAQ lists specialised startups, data providers (Mercor, Surge, Handshake, Turing), product companies and labs' in-house teams as sources ('Which companies build RL Environments?'); many alternatives to Handshake. Full date."}]},"switchingTime":{"level":2,"evidence":[{"url":"https://newsletter.semianalysis.com/p/rl-environments-and-rl-for-science","date":"2026-01-06","support":"SemiAnalysis says most environment vendors sell under exclusive contracts and UI-gym sites cost about US$20,000 each as one-time purchases (vendor and pricing sections); replacing Handshake's pipeline means commissioning new builds. Full date."},{"url":"https://techcrunch.com/2025/09/16/silicon-valley-bets-big-on-environments-to-train-ai-agents/","date":"2025-09-16","support":"Ross Taylor warns \"people are underestimating how difficult it is to scale environments\" (paragraph 32); a lab replacing Handshake's environment output would need months to re-commission it. Full date."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://epoch.ai/gradient-updates/state-of-rl-envs","date":"2026-01-12","support":"Epoch's RL-environment FAQ lists specialised startups, data providers (Mercor, Surge, Handshake, Turing), product companies and labs' in-house teams as sources ('Which companies build RL Environments?'); many alternatives to Handshake. Full date."}]}},"evaluation":{"alternatives":{"level":1,"evidence":[{"url":"https://techcrunch.com/2026/06/29/arena-the-ai-leaderboard-everyone-uses-is-now-a-100m-business/","date":"2026-06-29","support":"TechCrunch says human-labelling firms Mercor, Surge and Scale AI compete with Arena \"for the same dollar\" in evaluation (paragraph 7); Handshake's evaluation work has several qualified substitutes. Full date."},{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02-26","support":"Innodata's FY2025 10-K (Item 1, Competition) names Appen, CloudFactory, Surge AI, Invisible, Turing, Mercor, TELUS Digital and Scale AI as AI-data competitors; Handshake work has many qualified homes. Date: EDGAR filing date."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02-26","support":"Innodata's 10-K (Item 1A) says customer agreements are generally terminable on about 30-90 days' notice, including on selecting another provider; Handshake volume can be moved on that timescale. Date: EDGAR filing date."},{"url":"https://fortune.com/2025/06/19/openai-is-phasing-out-scale-ai-work-following-startups-meta-deal/","date":"2025-06-19","support":"OpenAI phased out Scale AI work within days of the Meta deal and relied increasingly on other providers such as Mercor (paragraphs 1, 9); Handshake's expert-data work is similarly portable. Full date."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://techcrunch.com/2026/06/29/arena-the-ai-leaderboard-everyone-uses-is-now-a-100m-business/","date":"2026-06-29","support":"TechCrunch says human-labelling firms Mercor, Surge and Scale AI compete with Arena \"for the same dollar\" in evaluation (paragraph 7); Handshake's evaluation work has several qualified substitutes. Full date."}]}}}},"b":{"schema":"5lc-index-rating/1","key":"models-data-evaluation","entityId":"handshake-ai","rater":"b","model":"claude-sonnet-5-5","ratedAt":"2026-10-22","functions":{"expert-data":{"alternatives":{"level":2,"evidence":[{"url":"https://time.com/collection/time100-ai/2026/garrett-lord/","date":"2026-08-27","support":"TIME reports Handshake AI works with eight frontier labs and names Scale AI, Surge and Mercor as its competitors (profile of CEO Garrett Lord, 27 Aug 2026)."},{"url":"https://techcrunch.com/2026/09/22/snorkel-ai-triples-valuation-to-3-5b-as-demand-for-ai-training-data-booms/","date":"2026-09-22","support":"Lists Mercor (about $2bn gross annualised), Handshake (about $1bn) and micro1 (about $500M gross) as similar AI-data companies; experts take roughly 60-70% of their top line."}]},"switchingTime":{"level":2,"evidence":[{"url":"https://time.com/collection/time100-ai/2026/garrett-lord/","date":"2026-08-27","support":"Handshake AI went from zero to over $1bn gross annualised in 16 months (company says), so a comparable vendor can be built or expanded to this volume in about a year (27 Aug 2026)."},{"url":"https://www.storyboard18.com/how-it-works/google-to-sever-ties-with-scale-ai-following-metas-significant-stake-70359.htm","date":"2025-06-16","support":"Google reportedly planned to end about $200M a year of Scale spend after Meta's stake; Handshake's CEO said demand 'tripled overnight' and Labelbox expected hundreds of millions in new revenue (report of 16 Jun 2025)."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://time.com/collection/time100-ai/2026/garrett-lord/","date":"2026-08-27","support":"TIME names Scale AI, Surge and Mercor as Handshake's competitors; Handshake's own edge is its 30M-user careers platform for recruiting experts (27 Aug 2026), not a capability others lack."},{"url":"https://techcrunch.com/2026/07/08/these-ai-startups-are-growing-revenue-at-faster-and-faster-rates/","date":"2026-07-08","support":"Mercor reports more than $2bn gross annualised (June 2026, company says), above Handshake's roughly $1bn."}]}},"rl-environments":{"alternatives":{"level":1,"evidence":[{"url":"https://newsletter.semianalysis.com/p/rl-environments-and-rl-for-science","date":"2026-01-06","support":"More than 35 companies sell RL environments (mostly seed-stage) and Anthropic likely works with more than a dozen of them (SemiAnalysis, 6 Jan 2026)."},{"url":"https://epoch.ai/gradient-updates/state-of-rl-envs","date":"2026-01-12","support":"Epoch lists specialist startups, the data vendors Mercor, Surge, Handshake and Turing, in-house lab teams and product-company partners as sources of environments (12 Jan 2026)."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://epoch.ai/gradient-updates/state-of-rl-envs","date":"2026-01-12","support":"Environment contracts are often six to seven figures per quarter and can be non-exclusive, so labs can add or swap vendors quarter by quarter; Epoch also reports 'substantially more' in-housing (12 Jan 2026)."},{"url":"https://newsletter.semianalysis.com/p/rl-environments-and-rl-for-science","date":"2026-01-06","support":"Anthropic is 'often the first customer' of new environment startups and wants a broad vendor ecosystem so the product is commoditised (SemiAnalysis, 6 Jan 2026)."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://epoch.ai/gradient-updates/state-of-rl-envs","date":"2026-01-12","support":"Quality at scale is the main bottleneck for all vendors; interviewees rank robustness to reward hacking first and say domain knowledge matters more than ML skill (12 Jan 2026). No vendor is described as uniquely capable."},{"url":"https://newsletter.semianalysis.com/p/rl-environments-and-rl-for-science","date":"2026-01-06","support":"Reports 'significant diversity in quality or focus' among vendors, with Turing on UI gyms and others on software engineering, rather than a single leader (6 Jan 2026)."}]}},"evaluation":{"alternatives":{"level":1,"evidence":[{"url":"https://labs.scale.com/leaderboard","date":"undated","support":"Scale Labs publishes 19 active leaderboards (page undated; the footer shows only a 2026 copyright)."},{"url":"https://www.prolific.com/","date":"undated","support":"Prolific sells human feedback for preference tuning, safety evals and benchmarks and lists a HUMAINE leaderboard (page undated; the footer shows only a 2026 copyright)."},{"url":"https://nebius.com/newsroom/nebius-welcomes-bezos-expeditions-as-lead-investor-in-ai-data-business-toloka","date":"2025-05-07","support":"Toloka offers agent safety evaluation and red teaming to labs (Nebius announcement, 7 May 2025)."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://mercor.com/blog/introducing-apex-ai-productivity-index/","date":"2025-10-01","support":"Mercor built its APEX v1 test set with about 100 experts in four professions (200 of 300 prompts approved), suggesting a comparable expert-written set is a matter of months (my inference, 1 Oct 2025)."},{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02","support":"Innodata says its customer agreements are generally terminable on 30 to 90 days' notice and do not commit customers to future purchases (risk factors). Date: month only."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://scale.com/blog/scales-next-era-building-for-2026","date":"2026-01-22","support":"Scale's lab says its lab introduced 15 new benchmarks and published over 450 evaluations of more than 50 models, a scale other vendors also reach (company blog, 22 Jan 2026)."},{"url":"https://welodata.ai/","date":"undated","support":"Welo Data lists evaluations, benchmarking, model selection and red teaming among its services (page undated; the footer shows only a 2026 copyright)."}]}}}}},"micro1":{"a":{"schema":"5lc-index-rating/1","key":"models-data-evaluation","entityId":"micro1","rater":"a","model":"claude-opus-5-5","ratedAt":"2026-10-22","functions":{"expert-data":{"alternatives":{"level":1,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02-26","support":"Innodata's FY2025 10-K (Item 1, Competition) names Appen, CloudFactory, Surge AI, Invisible, Turing, Mercor, TELUS Digital and Scale AI as AI-data competitors; micro1 work has many qualified homes. Date: EDGAR filing date."},{"url":"https://techcrunch.com/2026/09/22/snorkel-ai-triples-valuation-to-3-5b-as-demand-for-ai-training-data-booms/","date":"2026-09-22","support":"TechCrunch names Mercor (US$2bn gross), Handshake (US$1bn) and micro1 (US$500m) as scaled peers, and Snorkel at US$375m run-rate (company says) (paragraphs 4-5); several large alternatives to micro1. Full date."}]},"switchingTime":{"level":2,"evidence":[{"url":"https://techcrunch.com/2026/07/08/these-ai-startups-are-growing-revenue-at-faster-and-faster-rates/","date":"2026-07-08","support":"Mercor's CEO says it went from US$1bn to US$2bn gross annualised revenue in about four months to June 2026 (paragraph 5): the market re-staffs expert work for micro1-sized volumes within months. Full date."},{"url":"https://techcrunch.com/2026/08/20/ai-data-startup-micro1-reaches-500m-gross-run-rate-amid-ai-training-boom/","date":"2026-08-20","support":"micro1's own growth from US$100m to US$500m gross took eight months (paragraph 1): replacing that volume would also take months. Full date."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://techcrunch.com/2026/08/20/ai-data-startup-micro1-reaches-500m-gross-run-rate-amid-ai-training-boom/","date":"2026-08-20","support":"micro1 'still lags competitors like Mercor' (US$2bn) and Handshake (US$1bn) (paragraphs 7-8): larger alternatives exist. Full date."}]}},"rl-environments":{"alternatives":{"level":1,"evidence":[{"url":"https://newsletter.semianalysis.com/p/rl-environments-and-rl-for-science","date":"2026-01-06","support":"SemiAnalysis counts \"more than 35 companies\" building RL environments, mostly seed-stage, each serving 1-3 customers (vendor section); micro1 faces many substitutes. Full date."},{"url":"https://techcrunch.com/2026/08/20/ai-data-startup-micro1-reaches-500m-gross-run-rate-amid-ai-training-boom/","date":"2026-08-20","support":"micro1's experts build 'reinforcement learning gyms' (paragraph 4), as do many other vendors. Full date."}]},"switchingTime":{"level":2,"evidence":[{"url":"https://newsletter.semianalysis.com/p/rl-environments-and-rl-for-science","date":"2026-01-06","support":"SemiAnalysis says most environment vendors sell under exclusive contracts and UI-gym sites cost about US$20,000 each as one-time purchases (vendor and pricing sections); replacing micro1's pipeline means commissioning new builds. Full date."},{"url":"https://techcrunch.com/2025/09/16/silicon-valley-bets-big-on-environments-to-train-ai-agents/","date":"2025-09-16","support":"Ross Taylor warns \"people are underestimating how difficult it is to scale environments\" (paragraph 32); a lab replacing micro1's environment output would need months to re-commission it. Full date."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://epoch.ai/gradient-updates/state-of-rl-envs","date":"2026-01-12","support":"Epoch's RL-environment FAQ lists specialised startups, data providers (Mercor, Surge, Handshake, Turing), product companies and labs' in-house teams as sources ('Which companies build RL Environments?'); many alternatives to micro1. Full date."}]}},"evaluation":{"alternatives":{"level":1,"evidence":[{"url":"https://techcrunch.com/2026/06/29/arena-the-ai-leaderboard-everyone-uses-is-now-a-100m-business/","date":"2026-06-29","support":"TechCrunch says human-labelling firms Mercor, Surge and Scale AI compete with Arena \"for the same dollar\" in evaluation (paragraph 7); micro1's evaluation work has several qualified substitutes. Full date."},{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02-26","support":"Innodata's FY2025 10-K (Item 1, Competition) names Appen, CloudFactory, Surge AI, Invisible, Turing, Mercor, TELUS Digital and Scale AI as AI-data competitors; micro1 work has many qualified homes. Date: EDGAR filing date."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02-26","support":"Innodata's 10-K (Item 1A) says customer agreements are generally terminable on about 30-90 days' notice, including on selecting another provider; micro1 volume can be moved on that timescale. Date: EDGAR filing date."},{"url":"https://fortune.com/2025/06/19/openai-is-phasing-out-scale-ai-work-following-startups-meta-deal/","date":"2025-06-19","support":"OpenAI phased out Scale AI work within days of the Meta deal and relied increasingly on other providers such as Mercor (paragraphs 1, 9); micro1's expert-data work is similarly portable. Full date."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://techcrunch.com/2026/06/29/arena-the-ai-leaderboard-everyone-uses-is-now-a-100m-business/","date":"2026-06-29","support":"TechCrunch says human-labelling firms Mercor, Surge and Scale AI compete with Arena \"for the same dollar\" in evaluation (paragraph 7); micro1's evaluation work has several qualified substitutes. Full date."}]}}}},"b":{"schema":"5lc-index-rating/1","key":"models-data-evaluation","entityId":"micro1","rater":"b","model":"claude-sonnet-5-5","ratedAt":"2026-10-22","functions":{"expert-data":{"alternatives":{"level":2,"evidence":[{"url":"https://techcrunch.com/2026/09/22/snorkel-ai-triples-valuation-to-3-5b-as-demand-for-ai-training-data-booms/","date":"2026-09-22","support":"Lists Mercor (about $2bn gross annualised), Handshake (about $1bn) and micro1 (about $500M gross) as similar AI-data companies; experts take roughly 60-70% of their top line."},{"url":"https://www.micro1.ai/blog/q3","date":"2026-10-12","support":"micro1 says it has 'surpassed $1B annual run rate' (12 Oct 2026, company statement), a scale shared by only a handful of vendors."}]},"switchingTime":{"level":2,"evidence":[{"url":"https://www.micro1.ai/blog/q3","date":"2026-10-12","support":"micro1 says it passed a $500M run rate this summer and $1B by the end of Q3, showing a vendor can add hundreds of millions within months (12 Oct 2026, company statement)."},{"url":"https://www.storyboard18.com/how-it-works/google-to-sever-ties-with-scale-ai-following-metas-significant-stake-70359.htm","date":"2025-06-16","support":"Google reportedly planned to end about $200M a year of Scale spend after Meta's stake; Handshake's CEO said demand 'tripled overnight' and Labelbox expected hundreds of millions in new revenue (report of 16 Jun 2025)."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://www.micro1.ai/blog/q3","date":"2026-10-12","support":"micro1 says it has 'tens of thousands of experts' and about 180 staff; Mercor and Handshake report larger or equal run-rates, so alternatives show no capacity shortfall (12 Oct 2026)."},{"url":"https://techcrunch.com/2026/07/08/these-ai-startups-are-growing-revenue-at-faster-and-faster-rates/","date":"2026-07-08","support":"Mercor reports more than $2bn gross annualised (June 2026, company says)."}]}},"rl-environments":{"alternatives":{"level":1,"evidence":[{"url":"https://newsletter.semianalysis.com/p/rl-environments-and-rl-for-science","date":"2026-01-06","support":"More than 35 companies sell RL environments (mostly seed-stage) and Anthropic likely works with more than a dozen of them (SemiAnalysis, 6 Jan 2026)."},{"url":"https://epoch.ai/gradient-updates/state-of-rl-envs","date":"2026-01-12","support":"Epoch lists specialist startups, the data vendors Mercor, Surge, Handshake and Turing, in-house lab teams and product-company partners as sources of environments (12 Jan 2026)."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://epoch.ai/gradient-updates/state-of-rl-envs","date":"2026-01-12","support":"Environment contracts are often six to seven figures per quarter and can be non-exclusive, so labs can add or swap vendors quarter by quarter; Epoch also reports 'substantially more' in-housing (12 Jan 2026)."},{"url":"https://newsletter.semianalysis.com/p/rl-environments-and-rl-for-science","date":"2026-01-06","support":"Anthropic is 'often the first customer' of new environment startups and wants a broad vendor ecosystem so the product is commoditised (SemiAnalysis, 6 Jan 2026)."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://epoch.ai/gradient-updates/state-of-rl-envs","date":"2026-01-12","support":"Quality at scale is the main bottleneck for all vendors; interviewees rank robustness to reward hacking first and say domain knowledge matters more than ML skill (12 Jan 2026). No vendor is described as uniquely capable."},{"url":"https://newsletter.semianalysis.com/p/rl-environments-and-rl-for-science","date":"2026-01-06","support":"Reports 'significant diversity in quality or focus' among vendors, with Turing on UI gyms and others on software engineering, rather than a single leader (6 Jan 2026)."}]}},"evaluation":{"alternatives":{"level":1,"evidence":[{"url":"https://labs.scale.com/leaderboard","date":"undated","support":"Scale Labs publishes 19 active leaderboards (page undated; the footer shows only a 2026 copyright)."},{"url":"https://www.prolific.com/","date":"undated","support":"Prolific sells human feedback for preference tuning, safety evals and benchmarks and lists a HUMAINE leaderboard (page undated; the footer shows only a 2026 copyright)."},{"url":"https://nebius.com/newsroom/nebius-welcomes-bezos-expeditions-as-lead-investor-in-ai-data-business-toloka","date":"2025-05-07","support":"Toloka offers agent safety evaluation and red teaming to labs (Nebius announcement, 7 May 2025)."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://mercor.com/blog/introducing-apex-ai-productivity-index/","date":"2025-10-01","support":"Mercor built its APEX v1 test set with about 100 experts in four professions (200 of 300 prompts approved), suggesting a comparable expert-written set is a matter of months (my inference, 1 Oct 2025)."},{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02","support":"Innodata says its customer agreements are generally terminable on 30 to 90 days' notice and do not commit customers to future purchases (risk factors). Date: month only."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://scale.com/blog/scales-next-era-building-for-2026","date":"2026-01-22","support":"Scale's lab says its lab introduced 15 new benchmarks and published over 450 evaluations of more than 50 models, a scale other vendors also reach (company blog, 22 Jan 2026)."},{"url":"https://welodata.ai/","date":"undated","support":"Welo Data lists evaluations, benchmarking, model selection and red teaming among its services (page undated; the footer shows only a 2026 copyright)."}]}}}}},"scale-ai":{"a":{"schema":"5lc-index-rating/1","key":"models-data-evaluation","entityId":"scale-ai","rater":"a","model":"claude-opus-5-5","ratedAt":"2026-10-22","functions":{"expert-data":{"alternatives":{"level":1,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02-26","support":"Innodata's FY2025 10-K (Item 1, Competition) names Appen, CloudFactory, Surge AI, Invisible, Turing, Mercor, TELUS Digital and Scale AI as AI-data competitors; Scale AI work has many qualified homes. Date: EDGAR filing date."},{"url":"https://techcrunch.com/2026/09/22/snorkel-ai-triples-valuation-to-3-5b-as-demand-for-ai-training-data-booms/","date":"2026-09-22","support":"TechCrunch names Mercor (US$2bn gross), Handshake (US$1bn) and micro1 (US$500m) as scaled peers, and Snorkel at US$375m run-rate (company says) (paragraphs 4-5); several large alternatives to Scale AI. Full date."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://newsletter.semianalysis.com/p/rl-environments-and-rl-for-science","date":"2026-01-06","support":"After Meta's stake, 'multiple AI labs mostly stopped contracting Scale' (Scale section): labs replaced Scale's expert-data volume within weeks to months in 2025. Full date."},{"url":"https://fortune.com/2025/06/19/openai-is-phasing-out-scale-ai-work-following-startups-meta-deal/","date":"2025-06-19","support":"OpenAI phased out Scale AI work within days of the Meta deal and relied increasingly on other providers such as Mercor (paragraphs 1, 9); Scale AI's expert-data work is similarly portable. Full date."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://www.pymnts.com/artificial-intelligence-2/2025/surge-ai-reportedly-targeting-1-billion-in-first-ever-fundraise/","date":"2025-07-01","support":"Citing Reuters: Surge's 2024 revenue (over US$1bn) exceeded Scale's US$870m (paragraph 4): larger alternatives exist. Full date."}]}},"labelling-multimodal":{"alternatives":{"level":1,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02-26","support":"Innodata's FY2025 10-K (Item 1, Competition) names Appen, CloudFactory, Surge AI, Invisible, Turing, Mercor, TELUS Digital and Scale AI as AI-data competitors; Scale AI work has many qualified homes. Date: EDGAR filing date."},{"url":"https://www.sec.gov/Archives/edgar/data/1829864/000182986426000012/task-20251231.htm","date":"2026-03-05","support":"TaskUs's FY2025 10-K (Item 1, Our Competition) calls AI data services a \"large, rapidly changing, and fragmented global market\"; many vendors can take Scale AI's labelling work. Date: EDGAR filing date."}]},"switchingTime":{"level":2,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/1829864/000182986426000012/task-20251231.htm","date":"2026-03-05","support":"TaskUs's 10-K (Item 1A, labour risks) says hiring and training staff for increased demand \"takes time\"; absorbing Scale AI's large programmes would take months, not weeks. Date: EDGAR filing date."},{"url":"https://techcrunch.com/2025/06/14/google-reportedly-plans-to-cut-ties-with-scale-ai/","date":"2025-06-14","support":"Google, which planned to pay Scale US$200m in 2025, was talking to competitors about moving the work; OpenAI kept Scale \"as one of many vendors\" (paragraphs 2-3). Relevant to Scale AI. Full date."}]},"capabilityGap":{"level":2,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/1829864/000182986426000012/task-20251231.htm","date":"2026-03-05","support":"TaskUs's 10-K (Item 1) puts AI Services at 18% of US$1,183.5m 2025 revenue (about US$213m, 5LayerCake arithmetic), so rivals of Scale AI already run labelling at large scale. Date: EDGAR filing date."},{"url":"https://newsletter.semianalysis.com/p/rl-environments-and-rl-for-science","date":"2026-01-06","support":"SemiAnalysis puts Scale's 2024 revenue 'north of $1.4B' (Scale section); the largest labelling rivals are somewhat smaller, a small capacity gap at Scale's volume. Full date."}]}},"rl-environments":{"alternatives":{"level":1,"evidence":[{"url":"https://newsletter.semianalysis.com/p/rl-environments-and-rl-for-science","date":"2026-01-06","support":"SemiAnalysis counts \"more than 35 companies\" building RL environments, mostly seed-stage, each serving 1-3 customers (vendor section); Scale AI faces many substitutes. Full date."},{"url":"https://techcrunch.com/2025/09/16/silicon-valley-bets-big-on-environments-to-train-ai-agents/","date":"2025-09-16","support":"Scale has lost ground to competitors and is adapting to build environments (paragraph 19); many other builders already supply labs. Full date."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://techcrunch.com/2025/09/16/silicon-valley-bets-big-on-environments-to-train-ai-agents/","date":"2025-09-16","support":"TechCrunch says \"there's no shortage of startups\" supplying environments and that big labs also build them in-house (paragraphs 3-4); alternatives to Scale AI are plentiful. Full date."},{"url":"https://newsletter.semianalysis.com/p/rl-environments-and-rl-for-science","date":"2026-01-06","support":"SemiAnalysis says most environment vendors sell under exclusive contracts and UI-gym sites cost about US$20,000 each as one-time purchases (vendor and pricing sections); replacing Scale AI's pipeline means commissioning new builds. Full date."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://techcrunch.com/2026/09/22/snorkel-ai-triples-valuation-to-3-5b-as-demand-for-ai-training-data-booms/","date":"2026-09-22","support":"TechCrunch names Mercor (US$2bn gross), Handshake (US$1bn) and micro1 (US$500m) as scaled peers, and Snorkel at US$375m run-rate (company says) (paragraphs 4-5); several large alternatives to Scale AI. Full date."}]}},"evaluation":{"alternatives":{"level":1,"evidence":[{"url":"https://techcrunch.com/2026/06/29/arena-the-ai-leaderboard-everyone-uses-is-now-a-100m-business/","date":"2026-06-29","support":"TechCrunch says human-labelling firms Mercor, Surge and Scale AI compete with Arena \"for the same dollar\" in evaluation (paragraph 7); Scale AI's evaluation work has several qualified substitutes. Full date."},{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02-26","support":"Innodata's FY2025 10-K (Item 1, Competition) names Appen, CloudFactory, Surge AI, Invisible, Turing, Mercor, TELUS Digital and Scale AI as AI-data competitors; Scale AI work has many qualified homes. Date: EDGAR filing date."},{"url":"https://techcrunch.com/2026/06/29/arena-the-ai-leaderboard-everyone-uses-is-now-a-100m-business/","date":"2026-06-29","support":"Arena alone reports US$100m annualised evaluation revenue (paragraph 1), larger than Scale's evaluation share. Full date."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02-26","support":"Innodata's 10-K (Item 1A) says customer agreements are generally terminable on about 30-90 days' notice, including on selecting another provider; Scale AI volume can be moved on that timescale. Date: EDGAR filing date."},{"url":"https://fortune.com/2025/06/19/openai-is-phasing-out-scale-ai-work-following-startups-meta-deal/","date":"2025-06-19","support":"OpenAI phased out Scale AI work within days of the Meta deal and relied increasingly on other providers such as Mercor (paragraphs 1, 9); Scale AI's expert-data work is similarly portable. Full date."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://techcrunch.com/2026/06/29/arena-the-ai-leaderboard-everyone-uses-is-now-a-100m-business/","date":"2026-06-29","support":"TechCrunch says human-labelling firms Mercor, Surge and Scale AI compete with Arena \"for the same dollar\" in evaluation (paragraph 7); Scale AI's evaluation work has several qualified substitutes. Full date."}]}}}},"b":{"schema":"5lc-index-rating/1","key":"models-data-evaluation","entityId":"scale-ai","rater":"b","model":"claude-sonnet-5-5","ratedAt":"2026-10-22","functions":{"expert-data":{"alternatives":{"level":1,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02","support":"Competition section names nine AI-data specialists (including Appen, CloudFactory, Surge, Invisible, Turing, Mercor, TELUS Digital and Scale) plus large services firms; Scale is one of many. Date: month only, taken from the 20 Feb 2026 cover-page share count for the FY2025 filing."},{"url":"https://time.com/collection/time100-ai/2026/garrett-lord/","date":"2026-08-27","support":"Handshake AI says it works with eight frontier labs (27 Aug 2026), so several expert-data vendors are already qualified at labs."},{"url":"https://nebius.com/newsroom/nebius-welcomes-bezos-expeditions-as-lead-investor-in-ai-data-business-toloka","date":"2025-05-07","support":"Toloka lists Amazon, Anthropic, Microsoft, poolside, Recraft and Shopify as customers (7 May 2025), another vendor qualified at major buyers."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://www.storyboard18.com/how-it-works/google-to-sever-ties-with-scale-ai-following-metas-significant-stake-70359.htm","date":"2025-06-16","support":"Google reportedly planned to end about $200M a year of Scale spend after Meta's stake; Handshake's CEO said demand 'tripled overnight' and Labelbox expected hundreds of millions in new revenue (report of 16 Jun 2025)."},{"url":"https://siliconangle.com/2025/07/01/data-labeling-startup-surge-ai-seeking-1b-first-capital-raise-reports-say/","date":"2025-07-01","support":"Surge reportedly won former Scale customers including OpenAI and Google by 1 July 2025, under three weeks after Meta's stake was confirmed on 13 June (SiliconANGLE, citing reports)."},{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02","support":"Innodata says its customer agreements are generally terminable on 30 to 90 days' notice and do not commit customers to future purchases (risk factors). Date: month only."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://techcrunch.com/2026/07/08/these-ai-startups-are-growing-revenue-at-faster-and-faster-rates/","date":"2026-07-08","support":"Mercor says (CEO, TechCrunch 8 Jul 2026) gross annualised revenue rose from $500M (Sept 2025) to $1bn and then over $2bn (June 2026); Handshake and micro1 report about $1bn each (company statements), so several vendors already operate at the billion-dollar scale."},{"url":"https://newsletter.semianalysis.com/p/rl-environments-and-rl-for-science","date":"2026-01-06","support":"Multiple labs 'mostly stopped contracting' Scale after Meta's stake and the work was absorbed by others such as Surge; the article names no capability shortfall in the replacements (6 Jan 2026)."}]}},"labelling-multimodal":{"alternatives":{"level":1,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02","support":"Competition section names nine AI-data specialists (including Appen, CloudFactory, Surge, Invisible, Turing, Mercor, TELUS Digital and Scale) plus large services firms; Scale is one of many. Date: month only, taken from the 20 Feb 2026 cover-page share count for the FY2025 filing."},{"url":"https://www.sec.gov/Archives/edgar/data/0001829864/000182986426000147/earningsreleaseex991q22026.htm","date":"2026-08-05","support":"TaskUs reports AI Services as its fastest-growing line, up 25.8% year over year in Q2 2026, a large listed provider of AI data work (release of 5 Aug 2026)."},{"url":"https://announcements.asx.com.au/asxpdf/20260827/pdf/0739d4q7lyh3w2.pdf","date":"2026-08-27","support":"Appen China's annualised run-rate passed $175M in June 2026 and Appen Global is rebuilding; Appen is a further large vendor (ASX, 27 Aug 2026)."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://announcements.asx.com.au/asxpdf/20250226/pdf/06fyjdglv66466.pdf","date":"2025-02-26","support":"Google's termination of its Appen contract in Q1 FY24 cut Appen's FY24 group revenue 14.2% to $234.3M, showing a large buyer can withdraw a labelling programme within a quarter (ASX, 26 Feb 2025)."},{"url":"https://scale.com/blog/scales-next-era-building-for-2026","date":"2026-01-22","support":"Scale says its data business delivered over 150,000 hours of physical-AI and robotics data and onboarded 10 robotics customers in 2025; such customers are also served by TaskUs and others (22 Jan 2026)."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://www.investegate.co.uk/announcement/rns/rws-holdings--rws/half-year-financial-report/9612324","date":"2026-06-11","support":"RWS TrainAI revenue (Generate segment) grew 44.7% to GBP 99.4M in the half to March 2026, with a new global technology client won, so incumbents are adding programmes and volume has somewhere to go (11 Jun 2026)."},{"url":"https://www.sec.gov/Archives/edgar/data/0000868675/000110465926088975/tm2615221d2_ex99-1.htm","date":"2026-07-31","support":"TELUS Digital's AI and data solutions line is shrinking on client ramp-downs, which suggests spare capacity among incumbents (31 Jul 2026)."}]}},"rl-environments":{"alternatives":{"level":1,"evidence":[{"url":"https://newsletter.semianalysis.com/p/rl-environments-and-rl-for-science","date":"2026-01-06","support":"More than 35 companies sell RL environments (mostly seed-stage) and Anthropic likely works with more than a dozen of them (SemiAnalysis, 6 Jan 2026)."},{"url":"https://epoch.ai/gradient-updates/state-of-rl-envs","date":"2026-01-12","support":"Epoch lists specialist startups, the data vendors Mercor, Surge, Handshake and Turing, in-house lab teams and product-company partners as sources of environments (12 Jan 2026)."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://epoch.ai/gradient-updates/state-of-rl-envs","date":"2026-01-12","support":"Environment contracts are often six to seven figures per quarter and can be non-exclusive, so labs can add or swap vendors quarter by quarter; Epoch also reports 'substantially more' in-housing (12 Jan 2026)."},{"url":"https://newsletter.semianalysis.com/p/rl-environments-and-rl-for-science","date":"2026-01-06","support":"Anthropic is 'often the first customer' of new environment startups and wants a broad vendor ecosystem so the product is commoditised (SemiAnalysis, 6 Jan 2026)."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://epoch.ai/gradient-updates/state-of-rl-envs","date":"2026-01-12","support":"Quality at scale is the main bottleneck for all vendors; interviewees rank robustness to reward hacking first and say domain knowledge matters more than ML skill (12 Jan 2026). No vendor is described as uniquely capable."},{"url":"https://newsletter.semianalysis.com/p/rl-environments-and-rl-for-science","date":"2026-01-06","support":"Reports 'significant diversity in quality or focus' among vendors, with Turing on UI gyms and others on software engineering, rather than a single leader (6 Jan 2026)."}]}},"evaluation":{"alternatives":{"level":1,"evidence":[{"url":"https://labs.scale.com/leaderboard","date":"undated","support":"Scale's own leaderboards (19 active) sit beside Arena, Artificial Analysis, Mercor's APEX and Prolific's HUMAINE (page undated; the footer shows only a 2026 copyright)."},{"url":"https://mercor.com/blog/introducing-apex-ai-productivity-index/","date":"2025-10-01","support":"Mercor publishes its own APEX productivity benchmarks (1 Oct 2025)."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://mercor.com/blog/introducing-apex-ai-productivity-index/","date":"2025-10-01","support":"Mercor built its APEX v1 test set with about 100 experts in four professions (200 of 300 prompts approved), suggesting a comparable expert-written set is a matter of months (my inference, 1 Oct 2025)."},{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02","support":"Innodata says its customer agreements are generally terminable on 30 to 90 days' notice and do not commit customers to future purchases (risk factors). Date: month only."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://scale.com/blog/scales-next-era-building-for-2026","date":"2026-01-22","support":"Scale's lab says its lab introduced 15 new benchmarks and published over 450 evaluations of more than 50 models, a scale other vendors also reach (company blog, 22 Jan 2026)."},{"url":"https://welodata.ai/","date":"undated","support":"Welo Data lists evaluations, benchmarking, model selection and red teaming among its services (page undated; the footer shows only a 2026 copyright)."}]}}}}},"arena-ai":{"a":{"schema":"5lc-index-rating/1","key":"models-data-evaluation","entityId":"arena-ai","rater":"a","model":"claude-opus-5-5","ratedAt":"2026-10-22","functions":{"evaluation":{"alternatives":{"level":3,"evidence":[{"url":"https://techcrunch.com/2026/06/29/arena-the-ai-leaderboard-everyone-uses-is-now-a-100m-business/","date":"2026-06-29","support":"Arena says it has no direct competitors; crowdsourced rival Yupp shut down in March, while Mercor, Surge and Scale compete for the same dollar (paragraph 7). Full date."},{"url":"https://news.bgov.com/legal-ops-and-tech/scale-unveils-new-ai-model-leaderboard-system-to-rival-lmarena","date":"2025-09-22","support":"Scale launched SEAL Showdown, a public leaderboard to rival LMArena, with voters in 100+ countries (headline, paragraphs 1-2): one partial head-to-head alternative plus expert-evaluation services. Full date."}]},"switchingTime":{"level":3,"evidence":[{"url":"https://techcrunch.com/2026/06/29/arena-the-ai-leaderboard-everyone-uses-is-now-a-100m-business/","date":"2026-06-29","support":"Arena's leaderboard rests on over 10 million user evaluations (paragraph 2); building a comparable public vote base and comparison history would take a rival one to two years. Full date."},{"url":"https://www.latent.space/p/state-of-evals-lmarenas-17b-vision","date":"2026-01-06","support":"Arena reports 250M+ conversations, tens of millions a month, and 5M+ users (discussion section, 00:05:10): a large installed community to replicate. Full date."}]},"capabilityGap":{"level":3,"evidence":[{"url":"https://news.bgov.com/legal-ops-and-tech/scale-unveils-new-ai-model-leaderboard-system-to-rival-lmarena","date":"2025-09-22","support":"SEAL Showdown relies on unpaid volunteer contributors (paragraph 3); its vote volume is not disclosed, so the gap is judged material on thin evidence. Full date."},{"url":"https://techcrunch.com/2026/06/29/arena-the-ai-leaderboard-everyone-uses-is-now-a-100m-business/","date":"2026-06-29","support":"Yupp, the other crowdsourced model-comparison platform, shut down in March 2026 (paragraph 7), leaving no alternative at Arena's community scale. Full date."}]}}}},"b":{"schema":"5lc-index-rating/1","key":"models-data-evaluation","entityId":"arena-ai","rater":"b","model":"claude-sonnet-5-5","ratedAt":"2026-10-22","functions":{"evaluation":{"alternatives":{"level":3,"evidence":[{"url":"https://arena.ai/blog/arena-100m-revenue","date":"2026-06-29","support":"Arena says it has 82M+ human votes and 700M+ conversations and calls its data the largest human-preference dataset in the world (company claim, 29 Jun 2026); its blog names no competitor."},{"url":"https://artificialanalysis.ai/about","date":"undated","support":"Artificial Analysis (500+ models) and Scale Labs (19 leaderboards, see Scale page) publish independent benchmarks, but neither opened page shows an open human-vote system of Arena's scale; I count two to three partial alternatives (Artificial Analysis page undated)."}]},"switchingTime":{"level":3,"evidence":[{"url":"https://arxiv.org/abs/2504.20879","date":"2025-04-29","support":"The 'Leaderboard Illusion' paper shows Arena became the go-to leaderboard that labs privately test many variants on, with Google and OpenAI receiving about 19% and 20% of its data; a rival would need years of votes and lab trust (my inference; 29 Apr 2025)."},{"url":"https://arena.ai/blog/arena-100m-revenue","date":"2026-06-29","support":"Arena reached $100M annualised revenue within eight months of its enterprise launch, from a public community of 10M+ monthly visitors (29 Jun 2026)."}]},"capabilityGap":{"level":3,"evidence":[{"url":"https://arena.ai/blog/arena-100m-revenue","date":"2026-06-29","support":"Company claim of 82M+ votes and 10M+ monthly visitors; no alternative publishes a comparable human-vote count in the sources opened (29 Jun 2026; evidence thin)."},{"url":"https://arxiv.org/abs/2504.20879","date":"2025-04-29","support":"The authors estimate extra Arena data can give up to 112% relative gains on Arena itself (conservative estimate), which is why the votes are valuable to labs (29 Apr 2025)."}]}}}}},"snorkel-ai":{"a":{"schema":"5lc-index-rating/1","key":"models-data-evaluation","entityId":"snorkel-ai","rater":"a","model":"claude-opus-5-5","ratedAt":"2026-10-22","functions":{"expert-data":{"alternatives":{"level":1,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02-26","support":"Innodata's FY2025 10-K (Item 1, Competition) names Appen, CloudFactory, Surge AI, Invisible, Turing, Mercor, TELUS Digital and Scale AI as AI-data competitors; Snorkel AI work has many qualified homes. Date: EDGAR filing date."},{"url":"https://snorkel.ai/press/snorkel-ai-raises-100-million-to-build-better-evaluators-for-ai-models/","date":"2025-05-29","support":"Snorkel launched Expert Data-as-a-Service in May 2025 (paragraph 2), entering a market with many established vendors. Full date."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://fortune.com/2025/06/19/openai-is-phasing-out-scale-ai-work-following-startups-meta-deal/","date":"2025-06-19","support":"OpenAI phased out Scale AI work within days of the Meta deal and relied increasingly on other providers such as Mercor (paragraphs 1, 9); Snorkel AI's expert-data work is similarly portable. Full date."},{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02-26","support":"Innodata's 10-K (Item 1A) says customer agreements are generally terminable on about 30-90 days' notice, including on selecting another provider; Snorkel AI volume can be moved on that timescale. Date: EDGAR filing date."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://techcrunch.com/2026/09/22/snorkel-ai-triples-valuation-to-3-5b-as-demand-for-ai-training-data-booms/","date":"2026-09-22","support":"TechCrunch names Mercor (US$2bn gross), Handshake (US$1bn) and micro1 (US$500m) as scaled peers, and Snorkel at US$375m run-rate (company says) (paragraphs 4-5); several large alternatives to Snorkel AI. Full date."}]}},"rl-environments":{"alternatives":{"level":1,"evidence":[{"url":"https://newsletter.semianalysis.com/p/rl-environments-and-rl-for-science","date":"2026-01-06","support":"SemiAnalysis counts \"more than 35 companies\" building RL environments, mostly seed-stage, each serving 1-3 customers (vendor section); Snorkel AI faces many substitutes. Full date."},{"url":"https://techcrunch.com/2026/09/22/snorkel-ai-triples-valuation-to-3-5b-as-demand-for-ai-training-data-booms/","date":"2026-09-22","support":"TechCrunch says Snorkel sells RL environments and complete datasets (paragraph 6), alongside many specialist builders. Full date."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://techcrunch.com/2025/09/16/silicon-valley-bets-big-on-environments-to-train-ai-agents/","date":"2025-09-16","support":"TechCrunch says \"there's no shortage of startups\" supplying environments and that big labs also build them in-house (paragraphs 3-4); alternatives to Snorkel AI are plentiful. Full date."},{"url":"https://newsletter.semianalysis.com/p/rl-environments-and-rl-for-science","date":"2026-01-06","support":"SemiAnalysis says most environment vendors sell under exclusive contracts and UI-gym sites cost about US$20,000 each as one-time purchases (vendor and pricing sections); replacing Snorkel AI's pipeline means commissioning new builds. Full date."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://techcrunch.com/2026/09/22/snorkel-ai-triples-valuation-to-3-5b-as-demand-for-ai-training-data-booms/","date":"2026-09-22","support":"TechCrunch names Mercor (US$2bn gross), Handshake (US$1bn) and micro1 (US$500m) as scaled peers, and Snorkel at US$375m run-rate (company says) (paragraphs 4-5); several large alternatives to Snorkel AI. Full date."}]}},"evaluation":{"alternatives":{"level":1,"evidence":[{"url":"https://techcrunch.com/2026/06/29/arena-the-ai-leaderboard-everyone-uses-is-now-a-100m-business/","date":"2026-06-29","support":"TechCrunch says human-labelling firms Mercor, Surge and Scale AI compete with Arena \"for the same dollar\" in evaluation (paragraph 7); Snorkel AI's evaluation work has several qualified substitutes. Full date."},{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02-26","support":"Innodata's FY2025 10-K (Item 1, Competition) names Appen, CloudFactory, Surge AI, Invisible, Turing, Mercor, TELUS Digital and Scale AI as AI-data competitors; Snorkel AI work has many qualified homes. Date: EDGAR filing date."},{"url":"https://snorkel.ai/press/snorkel-ai-raises-100-million-to-build-better-evaluators-for-ai-models/","date":"2025-05-29","support":"Snorkel Evaluate launched in 2025 (paragraph 2), one of many expert-evaluation offerings. Full date."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02-26","support":"Innodata's 10-K (Item 1A) says customer agreements are generally terminable on about 30-90 days' notice, including on selecting another provider; Snorkel AI volume can be moved on that timescale. Date: EDGAR filing date."},{"url":"https://fortune.com/2025/06/19/openai-is-phasing-out-scale-ai-work-following-startups-meta-deal/","date":"2025-06-19","support":"OpenAI phased out Scale AI work within days of the Meta deal and relied increasingly on other providers such as Mercor (paragraphs 1, 9); Snorkel AI's expert-data work is similarly portable. Full date."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://techcrunch.com/2026/06/29/arena-the-ai-leaderboard-everyone-uses-is-now-a-100m-business/","date":"2026-06-29","support":"TechCrunch says human-labelling firms Mercor, Surge and Scale AI compete with Arena \"for the same dollar\" in evaluation (paragraph 7); Snorkel AI's evaluation work has several qualified substitutes. Full date."}]}},"data-tooling":{"alternatives":{"level":1,"evidence":[{"url":"https://lightly.ai/blog/the-10-best-encord-alternatives-in-2026-a-practical-guide-for-ml-team","date":"2026-05-18","support":"A May 2026 vendor guide lists ten labelling platforms (Labelbox, SuperAnnotate, Scale AI, V7, CVAT, Label Studio, Roboflow, Kili, Voxel51, Lightly) plus Dataloop as alternatives; many substitutes for Snorkel AI. Full date; author is a rival vendor."},{"url":"https://github.com/HumanSignal/label-studio","date":"undated","support":"Label Studio is an open-source, multi-type data labelling tool (text, audio, images, video, time series) with about 28.5k GitHub stars; a free substitute for Snorkel AI's software. Page undated."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://aws.amazon.com/about-aws/whats-new/2026/06/aws-service-availability/","date":"2026-06","support":"AWS ended SageMaker Ground Truth Plus on 30 June 2026 and closed Ground Truth to new customers from 30 July 2026: labelling-software users are expected to migrate within weeks. Relevant to Snorkel AI. Month taken from the URL path."},{"url":"https://docs.supervisely.com/import-and-export/migrations/migration-labelbox","date":"undated","support":"Supervisely offers a tool that copies multiple Labelbox projects into its platform; ready migration tooling keeps a switch away from Snorkel AI-type platforms to weeks. Page undated."},{"url":"https://lightly.ai/blog/the-10-best-encord-alternatives-in-2026-a-practical-guide-for-ml-team","date":"2026-05-18","support":"The guide advises \"Run the top two on a representative slice before you commit\", a short trial-based switch for platforms like Snorkel AI. Full date; author is a rival vendor."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://github.com/HumanSignal/label-studio","date":"undated","support":"Label Studio is an open-source, multi-type data labelling tool (text, audio, images, video, time series) with about 28.5k GitHub stars; a free substitute for Snorkel AI's software. Page undated."},{"url":"https://techcrunch.com/2026/09/22/snorkel-ai-triples-valuation-to-3-5b-as-demand-for-ai-training-data-booms/","date":"2026-09-22","support":"Snorkel began as data-labelling automation software (paragraph 3); open-source and commercial labelling tools offer comparable functions. Full date."}]}}}},"b":{"schema":"5lc-index-rating/1","key":"models-data-evaluation","entityId":"snorkel-ai","rater":"b","model":"claude-sonnet-5-5","ratedAt":"2026-10-22","functions":{"expert-data":{"alternatives":{"level":1,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02","support":"Competition section names nine AI-data specialists (including Appen, CloudFactory, Surge, Invisible, Turing, Mercor, TELUS Digital and Scale) plus large services firms; Snorkel is one of many. Date: month only, taken from the 20 Feb 2026 cover-page share count for the FY2025 filing."},{"url":"https://time.com/collection/time100-ai/2026/garrett-lord/","date":"2026-08-27","support":"Handshake AI says it works with eight frontier labs (27 Aug 2026), so several expert-data vendors are already qualified at labs."},{"url":"https://nebius.com/newsroom/nebius-welcomes-bezos-expeditions-as-lead-investor-in-ai-data-business-toloka","date":"2025-05-07","support":"Toloka lists Amazon, Anthropic, Microsoft, poolside, Recraft and Shopify as customers (7 May 2025), another vendor qualified at major buyers."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://www.storyboard18.com/how-it-works/google-to-sever-ties-with-scale-ai-following-metas-significant-stake-70359.htm","date":"2025-06-16","support":"Google reportedly planned to end about $200M a year of Scale spend after Meta's stake; Handshake's CEO said demand 'tripled overnight' and Labelbox expected hundreds of millions in new revenue (report of 16 Jun 2025)."},{"url":"https://siliconangle.com/2025/07/01/data-labeling-startup-surge-ai-seeking-1b-first-capital-raise-reports-say/","date":"2025-07-01","support":"Surge reportedly won former Scale customers including OpenAI and Google by 1 July 2025, under three weeks after Meta's stake was confirmed on 13 June (SiliconANGLE, citing reports)."},{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02","support":"Innodata says its customer agreements are generally terminable on 30 to 90 days' notice and do not commit customers to future purchases (risk factors). Date: month only."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://techcrunch.com/2026/07/08/these-ai-startups-are-growing-revenue-at-faster-and-faster-rates/","date":"2026-07-08","support":"Mercor says (CEO, TechCrunch 8 Jul 2026) gross annualised revenue rose from $500M (Sept 2025) to $1bn and then over $2bn (June 2026); Handshake and micro1 report about $1bn each (company statements), so several vendors already operate at the billion-dollar scale."},{"url":"https://newsletter.semianalysis.com/p/rl-environments-and-rl-for-science","date":"2026-01-06","support":"Multiple labs 'mostly stopped contracting' Scale after Meta's stake and the work was absorbed by others such as Surge; the article names no capability shortfall in the replacements (6 Jan 2026)."}]}},"rl-environments":{"alternatives":{"level":1,"evidence":[{"url":"https://newsletter.semianalysis.com/p/rl-environments-and-rl-for-science","date":"2026-01-06","support":"More than 35 companies sell RL environments (mostly seed-stage) and Anthropic likely works with more than a dozen of them (SemiAnalysis, 6 Jan 2026)."},{"url":"https://epoch.ai/gradient-updates/state-of-rl-envs","date":"2026-01-12","support":"Epoch lists specialist startups, the data vendors Mercor, Surge, Handshake and Turing, in-house lab teams and product-company partners as sources of environments (12 Jan 2026)."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://epoch.ai/gradient-updates/state-of-rl-envs","date":"2026-01-12","support":"Environment contracts are often six to seven figures per quarter and can be non-exclusive, so labs can add or swap vendors quarter by quarter; Epoch also reports 'substantially more' in-housing (12 Jan 2026)."},{"url":"https://newsletter.semianalysis.com/p/rl-environments-and-rl-for-science","date":"2026-01-06","support":"Anthropic is 'often the first customer' of new environment startups and wants a broad vendor ecosystem so the product is commoditised (SemiAnalysis, 6 Jan 2026)."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://epoch.ai/gradient-updates/state-of-rl-envs","date":"2026-01-12","support":"Quality at scale is the main bottleneck for all vendors; interviewees rank robustness to reward hacking first and say domain knowledge matters more than ML skill (12 Jan 2026). No vendor is described as uniquely capable."},{"url":"https://newsletter.semianalysis.com/p/rl-environments-and-rl-for-science","date":"2026-01-06","support":"Reports 'significant diversity in quality or focus' among vendors, with Turing on UI gyms and others on software engineering, rather than a single leader (6 Jan 2026)."}]}},"evaluation":{"alternatives":{"level":1,"evidence":[{"url":"https://labs.scale.com/leaderboard","date":"undated","support":"Scale Labs publishes 19 active leaderboards (page undated; the footer shows only a 2026 copyright)."},{"url":"https://www.prolific.com/","date":"undated","support":"Prolific sells human feedback for preference tuning, safety evals and benchmarks and lists a HUMAINE leaderboard (page undated; the footer shows only a 2026 copyright)."},{"url":"https://nebius.com/newsroom/nebius-welcomes-bezos-expeditions-as-lead-investor-in-ai-data-business-toloka","date":"2025-05-07","support":"Toloka offers agent safety evaluation and red teaming to labs (Nebius announcement, 7 May 2025)."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://mercor.com/blog/introducing-apex-ai-productivity-index/","date":"2025-10-01","support":"Mercor built its APEX v1 test set with about 100 experts in four professions (200 of 300 prompts approved), suggesting a comparable expert-written set is a matter of months (my inference, 1 Oct 2025)."},{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02","support":"Innodata says its customer agreements are generally terminable on 30 to 90 days' notice and do not commit customers to future purchases (risk factors). Date: month only."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://scale.com/blog/scales-next-era-building-for-2026","date":"2026-01-22","support":"Scale's lab says its lab introduced 15 new benchmarks and published over 450 evaluations of more than 50 models, a scale other vendors also reach (company blog, 22 Jan 2026)."},{"url":"https://welodata.ai/","date":"undated","support":"Welo Data lists evaluations, benchmarking, model selection and red teaming among its services (page undated; the footer shows only a 2026 copyright)."}]}},"data-tooling":{"alternatives":{"level":1,"evidence":[{"url":"https://techcrunch.com/2026/09/22/snorkel-ai-triples-valuation-to-3-5b-as-demand-for-ai-training-data-booms/","date":"2026-09-22","support":"Snorkel originally sold data-labelling automation software and shifted to data-as-a-service last year, i.e. software is no longer its main product (22 Sep 2026)."},{"url":"https://github.com/HumanSignal/label-studio","date":"undated","support":"Label Studio is an Apache-2.0 open-source labelling tool with about 28.5k GitHub stars covering images, audio, text, video and time series (page undated; the footer shows only a 2026 copyright)."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://www.vcaonline.com/news/2025071524/superannotate-raises-13-5-million-in-new-funding-from-dell-technologies-capital/","date":"2025-07-15","support":"SuperAnnotate says its software revenue grew 5x in 2024 with customers such as Databricks and Canva, i.e. buyers adopt a platform quickly (15 Jul 2025, company claim; evidence thin)."},{"url":"https://github.com/HumanSignal/label-studio","date":"undated","support":"Open-source tools can be self-hosted and imported to without licence or procurement, shortening any switch (page undated; the footer shows only a 2026 copyright)."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://snorkel.ai/blog/data-2-0-and-the-research-era-of-ai-data/","date":"2026-09-22","support":"Snorkel describes its agentic data platform as internal support for human experts, with growth coming from data-as-a-service (22 Sep 2026)."},{"url":"https://github.com/cvat-ai/cvat","date":"undated","support":"CVAT's MIT-licensed core already covers bounding boxes, polygons, masks, keypoints, cuboids and 3D point clouds (page undated; the footer shows only a 2026 copyright)."}]}}}}},"turing":{"a":{"schema":"5lc-index-rating/1","key":"models-data-evaluation","entityId":"turing","rater":"a","model":"claude-opus-5-5","ratedAt":"2026-10-22","functions":{"expert-data":{"alternatives":{"level":1,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02-26","support":"Innodata's FY2025 10-K (Item 1, Competition) names Appen, CloudFactory, Surge AI, Invisible, Turing, Mercor, TELUS Digital and Scale AI as AI-data competitors; Turing work has many qualified homes. Date: EDGAR filing date."},{"url":"https://www.turing.com/blog/turing-raises-111m-to-accelerate-the-future-of-agi","date":"2025-06-10","support":"Turing says it is expanding its role as a research accelerator for frontier labs (paragraph 4); it is one of many such vendors. Full date."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://fortune.com/2025/06/19/openai-is-phasing-out-scale-ai-work-following-startups-meta-deal/","date":"2025-06-19","support":"OpenAI phased out Scale AI work within days of the Meta deal and relied increasingly on other providers such as Mercor (paragraphs 1, 9); Turing's expert-data work is similarly portable. Full date."},{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02-26","support":"Innodata's 10-K (Item 1A) says customer agreements are generally terminable on about 30-90 days' notice, including on selecting another provider; Turing volume can be moved on that timescale. Date: EDGAR filing date."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://techcrunch.com/2026/09/22/snorkel-ai-triples-valuation-to-3-5b-as-demand-for-ai-training-data-booms/","date":"2026-09-22","support":"TechCrunch names Mercor (US$2bn gross), Handshake (US$1bn) and micro1 (US$500m) as scaled peers, and Snorkel at US$375m run-rate (company says) (paragraphs 4-5); several large alternatives to Turing. Full date."}]}},"rl-environments":{"alternatives":{"level":1,"evidence":[{"url":"https://newsletter.semianalysis.com/p/rl-environments-and-rl-for-science","date":"2026-01-06","support":"SemiAnalysis counts \"more than 35 companies\" building RL environments, mostly seed-stage, each serving 1-3 customers (vendor section); Turing faces many substitutes. Full date."},{"url":"https://newsletter.semianalysis.com/p/rl-environments-and-rl-for-science","date":"2026-01-06","support":"SemiAnalysis notes diversity of focus, with Turing on UI gyms and Mechanize on software-engineering tasks (quality section); other builders cover the same work. Full date."}]},"switchingTime":{"level":2,"evidence":[{"url":"https://newsletter.semianalysis.com/p/rl-environments-and-rl-for-science","date":"2026-01-06","support":"SemiAnalysis says most environment vendors sell under exclusive contracts and UI-gym sites cost about US$20,000 each as one-time purchases (vendor and pricing sections); replacing Turing's pipeline means commissioning new builds. Full date."},{"url":"https://techcrunch.com/2025/09/16/silicon-valley-bets-big-on-environments-to-train-ai-agents/","date":"2025-09-16","support":"Ross Taylor warns \"people are underestimating how difficult it is to scale environments\" (paragraph 32); a lab replacing Turing's environment output would need months to re-commission it. Full date."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://epoch.ai/gradient-updates/state-of-rl-envs","date":"2026-01-12","support":"Epoch's RL-environment FAQ lists specialised startups, data providers (Mercor, Surge, Handshake, Turing), product companies and labs' in-house teams as sources ('Which companies build RL Environments?'); many alternatives to Turing. Full date."}]}},"evaluation":{"alternatives":{"level":1,"evidence":[{"url":"https://techcrunch.com/2026/06/29/arena-the-ai-leaderboard-everyone-uses-is-now-a-100m-business/","date":"2026-06-29","support":"TechCrunch says human-labelling firms Mercor, Surge and Scale AI compete with Arena \"for the same dollar\" in evaluation (paragraph 7); Turing's evaluation work has several qualified substitutes. Full date."},{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02-26","support":"Innodata's FY2025 10-K (Item 1, Competition) names Appen, CloudFactory, Surge AI, Invisible, Turing, Mercor, TELUS Digital and Scale AI as AI-data competitors; Turing work has many qualified homes. Date: EDGAR filing date."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02-26","support":"Innodata's 10-K (Item 1A) says customer agreements are generally terminable on about 30-90 days' notice, including on selecting another provider; Turing volume can be moved on that timescale. Date: EDGAR filing date."},{"url":"https://fortune.com/2025/06/19/openai-is-phasing-out-scale-ai-work-following-startups-meta-deal/","date":"2025-06-19","support":"OpenAI phased out Scale AI work within days of the Meta deal and relied increasingly on other providers such as Mercor (paragraphs 1, 9); Turing's expert-data work is similarly portable. Full date."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://techcrunch.com/2026/06/29/arena-the-ai-leaderboard-everyone-uses-is-now-a-100m-business/","date":"2026-06-29","support":"TechCrunch says human-labelling firms Mercor, Surge and Scale AI compete with Arena \"for the same dollar\" in evaluation (paragraph 7); Turing's evaluation work has several qualified substitutes. Full date."}]}}}},"b":{"schema":"5lc-index-rating/1","key":"models-data-evaluation","entityId":"turing","rater":"b","model":"claude-sonnet-5-5","ratedAt":"2026-10-22","functions":{"expert-data":{"alternatives":{"level":1,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02","support":"Competition section names nine AI-data specialists (including Appen, CloudFactory, Surge, Invisible, Turing, Mercor, TELUS Digital and Scale) plus large services firms; Turing is one of many. Date: month only, taken from the 20 Feb 2026 cover-page share count for the FY2025 filing."},{"url":"https://time.com/collection/time100-ai/2026/garrett-lord/","date":"2026-08-27","support":"Handshake AI says it works with eight frontier labs (27 Aug 2026), so several expert-data vendors are already qualified at labs."},{"url":"https://nebius.com/newsroom/nebius-welcomes-bezos-expeditions-as-lead-investor-in-ai-data-business-toloka","date":"2025-05-07","support":"Toloka lists Amazon, Anthropic, Microsoft, poolside, Recraft and Shopify as customers (7 May 2025), another vendor qualified at major buyers."},{"url":"https://techcrunch.com/2025/03/06/turing-a-key-coding-provider-for-openai-and-other-llm-producers-raises-111m-at-a-2-2b-valuation","date":"2025-03-06","support":"Turing's lab work is largely coding data for OpenAI and other model makers (TechCrunch, 6 Mar 2025); Mercor, Handshake and Surge also hire engineers as domain experts for labs (SemiAnalysis, Jan 2026)."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://www.storyboard18.com/how-it-works/google-to-sever-ties-with-scale-ai-following-metas-significant-stake-70359.htm","date":"2025-06-16","support":"Google reportedly planned to end about $200M a year of Scale spend after Meta's stake; Handshake's CEO said demand 'tripled overnight' and Labelbox expected hundreds of millions in new revenue (report of 16 Jun 2025)."},{"url":"https://siliconangle.com/2025/07/01/data-labeling-startup-surge-ai-seeking-1b-first-capital-raise-reports-say/","date":"2025-07-01","support":"Surge reportedly won former Scale customers including OpenAI and Google by 1 July 2025, under three weeks after Meta's stake was confirmed on 13 June (SiliconANGLE, citing reports)."},{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02","support":"Innodata says its customer agreements are generally terminable on 30 to 90 days' notice and do not commit customers to future purchases (risk factors). Date: month only."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://techcrunch.com/2026/07/08/these-ai-startups-are-growing-revenue-at-faster-and-faster-rates/","date":"2026-07-08","support":"Mercor says (CEO, TechCrunch 8 Jul 2026) gross annualised revenue rose from $500M (Sept 2025) to $1bn and then over $2bn (June 2026); Handshake and micro1 report about $1bn each (company statements), so several vendors already operate at the billion-dollar scale."},{"url":"https://newsletter.semianalysis.com/p/rl-environments-and-rl-for-science","date":"2026-01-06","support":"Multiple labs 'mostly stopped contracting' Scale after Meta's stake and the work was absorbed by others such as Surge; the article names no capability shortfall in the replacements (6 Jan 2026)."}]}},"rl-environments":{"alternatives":{"level":1,"evidence":[{"url":"https://newsletter.semianalysis.com/p/rl-environments-and-rl-for-science","date":"2026-01-06","support":"More than 35 companies sell RL environments (mostly seed-stage) and Anthropic likely works with more than a dozen of them (SemiAnalysis, 6 Jan 2026)."},{"url":"https://epoch.ai/gradient-updates/state-of-rl-envs","date":"2026-01-12","support":"Epoch lists specialist startups, the data vendors Mercor, Surge, Handshake and Turing, in-house lab teams and product-company partners as sources of environments (12 Jan 2026)."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://epoch.ai/gradient-updates/state-of-rl-envs","date":"2026-01-12","support":"Environment contracts are often six to seven figures per quarter and can be non-exclusive, so labs can add or swap vendors quarter by quarter; Epoch also reports 'substantially more' in-housing (12 Jan 2026)."},{"url":"https://newsletter.semianalysis.com/p/rl-environments-and-rl-for-science","date":"2026-01-06","support":"Anthropic is 'often the first customer' of new environment startups and wants a broad vendor ecosystem so the product is commoditised (SemiAnalysis, 6 Jan 2026)."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://newsletter.semianalysis.com/p/rl-environments-and-rl-for-science","date":"2026-01-06","support":"Turing is named as a UI-gym builder among many environment vendors with 'significant diversity in quality or focus' (6 Jan 2026)."},{"url":"https://epoch.ai/gradient-updates/state-of-rl-envs","date":"2026-01-12","support":"Quality at scale is the main bottleneck for all vendors; interviewees rank robustness to reward hacking first and say domain knowledge matters more than ML skill (12 Jan 2026). No vendor is described as uniquely capable."}]}},"evaluation":{"alternatives":{"level":1,"evidence":[{"url":"https://labs.scale.com/leaderboard","date":"undated","support":"Scale Labs publishes 19 active leaderboards (page undated; the footer shows only a 2026 copyright)."},{"url":"https://www.prolific.com/","date":"undated","support":"Prolific sells human feedback for preference tuning, safety evals and benchmarks and lists a HUMAINE leaderboard (page undated; the footer shows only a 2026 copyright)."},{"url":"https://nebius.com/newsroom/nebius-welcomes-bezos-expeditions-as-lead-investor-in-ai-data-business-toloka","date":"2025-05-07","support":"Toloka offers agent safety evaluation and red teaming to labs (Nebius announcement, 7 May 2025)."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://mercor.com/blog/introducing-apex-ai-productivity-index/","date":"2025-10-01","support":"Mercor built its APEX v1 test set with about 100 experts in four professions (200 of 300 prompts approved), suggesting a comparable expert-written set is a matter of months (my inference, 1 Oct 2025)."},{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02","support":"Innodata says its customer agreements are generally terminable on 30 to 90 days' notice and do not commit customers to future purchases (risk factors). Date: month only."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://scale.com/blog/scales-next-era-building-for-2026","date":"2026-01-22","support":"Scale's lab says its lab introduced 15 new benchmarks and published over 450 evaluations of more than 50 models, a scale other vendors also reach (company blog, 22 Jan 2026)."},{"url":"https://welodata.ai/","date":"undated","support":"Welo Data lists evaluations, benchmarking, model selection and red teaming among its services (page undated; the footer shows only a 2026 copyright)."}]}}}}},"innodata":{"a":{"schema":"5lc-index-rating/1","key":"models-data-evaluation","entityId":"innodata","rater":"a","model":"claude-opus-5-5","ratedAt":"2026-10-22","functions":{"expert-data":{"alternatives":{"level":1,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02-26","support":"Innodata's FY2025 10-K (Item 1, Competition) names Appen, CloudFactory, Surge AI, Invisible, Turing, Mercor, TELUS Digital and Scale AI as AI-data competitors; Innodata work has many qualified homes. Date: EDGAR filing date."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://fortune.com/2025/06/19/openai-is-phasing-out-scale-ai-work-following-startups-meta-deal/","date":"2025-06-19","support":"OpenAI phased out Scale AI work within days of the Meta deal and relied increasingly on other providers such as Mercor (paragraphs 1, 9); Innodata's expert-data work is similarly portable. Full date."},{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02-26","support":"Innodata's 10-K (Item 1A) says customer agreements are generally terminable on about 30-90 days' notice, including on selecting another provider; Innodata volume can be moved on that timescale. Date: EDGAR filing date."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://techcrunch.com/2026/09/22/snorkel-ai-triples-valuation-to-3-5b-as-demand-for-ai-training-data-booms/","date":"2026-09-22","support":"TechCrunch names Mercor (US$2bn gross), Handshake (US$1bn) and micro1 (US$500m) as scaled peers, and Snorkel at US$375m run-rate (company says) (paragraphs 4-5); several large alternatives to Innodata. Full date."}]}},"labelling-multimodal":{"alternatives":{"level":1,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02-26","support":"Innodata's FY2025 10-K (Item 1, Competition) names Appen, CloudFactory, Surge AI, Invisible, Turing, Mercor, TELUS Digital and Scale AI as AI-data competitors; Innodata work has many qualified homes. Date: EDGAR filing date."},{"url":"https://www.sec.gov/Archives/edgar/data/1829864/000182986426000012/task-20251231.htm","date":"2026-03-05","support":"TaskUs's FY2025 10-K (Item 1, Our Competition) calls AI data services a \"large, rapidly changing, and fragmented global market\"; many vendors can take Innodata's labelling work. Date: EDGAR filing date."},{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02-26","support":"Innodata's 10-K (Item 1, Our Customers) says one customer was about 58% of 2025 revenue; that customer also uses other vendors such as those Innodata lists as competitors. Date: EDGAR filing date."}]},"switchingTime":{"level":2,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/1829864/000182986426000012/task-20251231.htm","date":"2026-03-05","support":"TaskUs's 10-K (Item 1A, labour risks) says hiring and training staff for increased demand \"takes time\"; absorbing Innodata's large programmes would take months, not weeks. Date: EDGAR filing date."},{"url":"https://www.searchenginejournal.com/google-terminates-contract-with-appen-for-search-quality-raters/506006/","date":"2024-01-21","support":"Google gave Appen notice on 20 January 2024 and ended its services by 19 March 2024 (paragraphs 1-2), a two-month exit; Innodata's volume could move similarly. Full date."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/1829864/000182986426000012/task-20251231.htm","date":"2026-03-05","support":"TaskUs's 10-K (Item 1) puts AI Services at 18% of US$1,183.5m 2025 revenue (about US$213m, 5LayerCake arithmetic), so rivals of Innodata already run labelling at large scale. Date: EDGAR filing date."}]}},"evaluation":{"alternatives":{"level":1,"evidence":[{"url":"https://techcrunch.com/2026/06/29/arena-the-ai-leaderboard-everyone-uses-is-now-a-100m-business/","date":"2026-06-29","support":"TechCrunch says human-labelling firms Mercor, Surge and Scale AI compete with Arena \"for the same dollar\" in evaluation (paragraph 7); Innodata's evaluation work has several qualified substitutes. Full date."},{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02-26","support":"Innodata's FY2025 10-K (Item 1, Competition) names Appen, CloudFactory, Surge AI, Invisible, Turing, Mercor, TELUS Digital and Scale AI as AI-data competitors; Innodata work has many qualified homes. Date: EDGAR filing date."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02-26","support":"Innodata's 10-K (Item 1A) says customer agreements are generally terminable on about 30-90 days' notice, including on selecting another provider; Innodata volume can be moved on that timescale. Date: EDGAR filing date."},{"url":"https://fortune.com/2025/06/19/openai-is-phasing-out-scale-ai-work-following-startups-meta-deal/","date":"2025-06-19","support":"OpenAI phased out Scale AI work within days of the Meta deal and relied increasingly on other providers such as Mercor (paragraphs 1, 9); Innodata's expert-data work is similarly portable. Full date."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://techcrunch.com/2026/06/29/arena-the-ai-leaderboard-everyone-uses-is-now-a-100m-business/","date":"2026-06-29","support":"TechCrunch says human-labelling firms Mercor, Surge and Scale AI compete with Arena \"for the same dollar\" in evaluation (paragraph 7); Innodata's evaluation work has several qualified substitutes. Full date."}]}}}},"b":{"schema":"5lc-index-rating/1","key":"models-data-evaluation","entityId":"innodata","rater":"b","model":"claude-sonnet-5-5","ratedAt":"2026-10-22","functions":{"expert-data":{"alternatives":{"level":1,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02","support":"Competition section names nine AI-data specialists (including Appen, CloudFactory, Surge, Invisible, Turing, Mercor, TELUS Digital and Scale) plus large services firms; Innodata is one of many. Date: month only, taken from the 20 Feb 2026 cover-page share count for the FY2025 filing."},{"url":"https://time.com/collection/time100-ai/2026/garrett-lord/","date":"2026-08-27","support":"Handshake AI says it works with eight frontier labs (27 Aug 2026), so several expert-data vendors are already qualified at labs."},{"url":"https://nebius.com/newsroom/nebius-welcomes-bezos-expeditions-as-lead-investor-in-ai-data-business-toloka","date":"2025-05-07","support":"Toloka lists Amazon, Anthropic, Microsoft, poolside, Recraft and Shopify as customers (7 May 2025), another vendor qualified at major buyers."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://www.storyboard18.com/how-it-works/google-to-sever-ties-with-scale-ai-following-metas-significant-stake-70359.htm","date":"2025-06-16","support":"Google reportedly planned to end about $200M a year of Scale spend after Meta's stake; Handshake's CEO said demand 'tripled overnight' and Labelbox expected hundreds of millions in new revenue (report of 16 Jun 2025)."},{"url":"https://siliconangle.com/2025/07/01/data-labeling-startup-surge-ai-seeking-1b-first-capital-raise-reports-say/","date":"2025-07-01","support":"Surge reportedly won former Scale customers including OpenAI and Google by 1 July 2025, under three weeks after Meta's stake was confirmed on 13 June (SiliconANGLE, citing reports)."},{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02","support":"Innodata says its customer agreements are generally terminable on 30 to 90 days' notice and do not commit customers to future purchases (risk factors). Date: month only."},{"url":"https://www.sec.gov/Archives/edgar/data/0000903651/000110465926092010/tm2621499d1_ex99-1.htm","date":"2026-08-06","support":"Innodata's largest customer fell from 56% of revenue in Q1 to 37% in Q2 2026 while another Big Tech customer rose to 34%, showing volume moves among customers fast (6 Aug 2026)."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://techcrunch.com/2026/07/08/these-ai-startups-are-growing-revenue-at-faster-and-faster-rates/","date":"2026-07-08","support":"Mercor says (CEO, TechCrunch 8 Jul 2026) gross annualised revenue rose from $500M (Sept 2025) to $1bn and then over $2bn (June 2026); Handshake and micro1 report about $1bn each (company statements), so several vendors already operate at the billion-dollar scale."},{"url":"https://newsletter.semianalysis.com/p/rl-environments-and-rl-for-science","date":"2026-01-06","support":"Multiple labs 'mostly stopped contracting' Scale after Meta's stake and the work was absorbed by others such as Surge; the article names no capability shortfall in the replacements (6 Jan 2026)."}]}},"labelling-multimodal":{"alternatives":{"level":1,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02","support":"Competition section names nine AI-data specialists (including Appen, CloudFactory, Surge, Invisible, Turing, Mercor, TELUS Digital and Scale) plus large services firms; Innodata is one of many. Date: month only, taken from the 20 Feb 2026 cover-page share count for the FY2025 filing."},{"url":"https://www.sec.gov/Archives/edgar/data/0001829864/000182986426000147/earningsreleaseex991q22026.htm","date":"2026-08-05","support":"TaskUs reports AI Services as its fastest-growing line, up 25.8% year over year in Q2 2026, a large listed provider of AI data work (release of 5 Aug 2026)."},{"url":"https://announcements.asx.com.au/asxpdf/20260827/pdf/0739d4q7lyh3w2.pdf","date":"2026-08-27","support":"Appen China's annualised run-rate passed $175M in June 2026 and Appen Global is rebuilding; Appen is a further large vendor (ASX, 27 Aug 2026)."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://announcements.asx.com.au/asxpdf/20250226/pdf/06fyjdglv66466.pdf","date":"2025-02-26","support":"Google's termination of its Appen contract in Q1 FY24 cut Appen's FY24 group revenue 14.2% to $234.3M, showing a large buyer can withdraw a labelling programme within a quarter (ASX, 26 Feb 2025)."},{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02","support":"Innodata says its customer agreements are generally terminable on 30 to 90 days' notice and do not commit customers to future purchases (risk factors). Date: month only."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://www.investegate.co.uk/announcement/rns/rws-holdings--rws/half-year-financial-report/9612324","date":"2026-06-11","support":"RWS TrainAI revenue (Generate segment) grew 44.7% to GBP 99.4M in the half to March 2026, with a new global technology client won, so incumbents are adding programmes and volume has somewhere to go (11 Jun 2026)."},{"url":"https://www.sec.gov/Archives/edgar/data/0000868675/000110465926088975/tm2615221d2_ex99-1.htm","date":"2026-07-31","support":"TELUS Digital's AI and data solutions line is shrinking on client ramp-downs, which suggests spare capacity among incumbents (31 Jul 2026)."}]}},"evaluation":{"alternatives":{"level":1,"evidence":[{"url":"https://labs.scale.com/leaderboard","date":"undated","support":"Scale Labs publishes 19 active leaderboards (page undated; the footer shows only a 2026 copyright)."},{"url":"https://www.prolific.com/","date":"undated","support":"Prolific sells human feedback for preference tuning, safety evals and benchmarks and lists a HUMAINE leaderboard (page undated; the footer shows only a 2026 copyright)."},{"url":"https://nebius.com/newsroom/nebius-welcomes-bezos-expeditions-as-lead-investor-in-ai-data-business-toloka","date":"2025-05-07","support":"Toloka offers agent safety evaluation and red teaming to labs (Nebius announcement, 7 May 2025)."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://mercor.com/blog/introducing-apex-ai-productivity-index/","date":"2025-10-01","support":"Mercor built its APEX v1 test set with about 100 experts in four professions (200 of 300 prompts approved), suggesting a comparable expert-written set is a matter of months (my inference, 1 Oct 2025)."},{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02","support":"Innodata says its customer agreements are generally terminable on 30 to 90 days' notice and do not commit customers to future purchases (risk factors). Date: month only."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://scale.com/blog/scales-next-era-building-for-2026","date":"2026-01-22","support":"Scale's lab says its lab introduced 15 new benchmarks and published over 450 evaluations of more than 50 models, a scale other vendors also reach (company blog, 22 Jan 2026)."},{"url":"https://welodata.ai/","date":"undated","support":"Welo Data lists evaluations, benchmarking, model selection and red teaming among its services (page undated; the footer shows only a 2026 copyright)."}]}}}}},"prolific":{"a":{"schema":"5lc-index-rating/1","key":"models-data-evaluation","entityId":"prolific","rater":"a","model":"claude-opus-5-5","ratedAt":"2026-10-22","functions":{"expert-data":{"alternatives":{"level":1,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02-26","support":"Innodata's FY2025 10-K (Item 1, Competition) names Appen, CloudFactory, Surge AI, Invisible, Turing, Mercor, TELUS Digital and Scale AI as AI-data competitors; Prolific work has many qualified homes. Date: EDGAR filing date."},{"url":"https://www.prolific.com/resources/prolific-in-2025-more-precision-larger-scale-and-stronger-safeguards-for-quality-human-data","date":"2025-12-16","support":"Prolific says frontier labs needed verified experts and it now runs full AI training projects (paragraphs 3, 9), competing with many expert-data vendors. Full date."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://fortune.com/2025/06/19/openai-is-phasing-out-scale-ai-work-following-startups-meta-deal/","date":"2025-06-19","support":"OpenAI phased out Scale AI work within days of the Meta deal and relied increasingly on other providers such as Mercor (paragraphs 1, 9); Prolific's expert-data work is similarly portable. Full date."},{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02-26","support":"Innodata's 10-K (Item 1A) says customer agreements are generally terminable on about 30-90 days' notice, including on selecting another provider; Prolific volume can be moved on that timescale. Date: EDGAR filing date."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://techcrunch.com/2026/09/22/snorkel-ai-triples-valuation-to-3-5b-as-demand-for-ai-training-data-booms/","date":"2026-09-22","support":"TechCrunch names Mercor (US$2bn gross), Handshake (US$1bn) and micro1 (US$500m) as scaled peers, and Snorkel at US$375m run-rate (company says) (paragraphs 4-5); several large alternatives to Prolific. Full date."}]}},"labelling-multimodal":{"alternatives":{"level":1,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02-26","support":"Innodata's FY2025 10-K (Item 1, Competition) names Appen, CloudFactory, Surge AI, Invisible, Turing, Mercor, TELUS Digital and Scale AI as AI-data competitors; Prolific work has many qualified homes. Date: EDGAR filing date."},{"url":"https://www.sec.gov/Archives/edgar/data/1829864/000182986426000012/task-20251231.htm","date":"2026-03-05","support":"TaskUs's FY2025 10-K (Item 1, Our Competition) calls AI data services a \"large, rapidly changing, and fragmented global market\"; many vendors can take Prolific's labelling work. Date: EDGAR filing date."},{"url":"https://www.prolific.com/resources/prolific-in-2025-more-precision-larger-scale-and-stronger-safeguards-for-quality-human-data","date":"2025-12-16","support":"Prolific reports 380,000 studies and tasks and 11,000 verified annotation specialists in 2025 (paragraph 2, specialist pools), capacity rivals also hold. Full date."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://www.searchenginejournal.com/google-terminates-contract-with-appen-for-search-quality-raters/506006/","date":"2024-01-21","support":"Google gave Appen notice on 20 January 2024 and ended its services by 19 March 2024 (paragraphs 1-2), a two-month exit; Prolific's volume could move similarly. Full date."},{"url":"https://www.sec.gov/Archives/edgar/data/1829864/000182986426000012/task-20251231.htm","date":"2026-03-05","support":"TaskUs's 10-K (Item 1A) says clients can terminate for convenience or move work in-house or to competitors; Prolific's programmes are portable within weeks to months. Date: EDGAR filing date."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/1829864/000182986426000012/task-20251231.htm","date":"2026-03-05","support":"TaskUs's 10-K (Item 1) puts AI Services at 18% of US$1,183.5m 2025 revenue (about US$213m, 5LayerCake arithmetic), so rivals of Prolific already run labelling at large scale. Date: EDGAR filing date."}]}},"evaluation":{"alternatives":{"level":1,"evidence":[{"url":"https://techcrunch.com/2026/06/29/arena-the-ai-leaderboard-everyone-uses-is-now-a-100m-business/","date":"2026-06-29","support":"TechCrunch says human-labelling firms Mercor, Surge and Scale AI compete with Arena \"for the same dollar\" in evaluation (paragraph 7); Prolific's evaluation work has several qualified substitutes. Full date."},{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02-26","support":"Innodata's FY2025 10-K (Item 1, Competition) names Appen, CloudFactory, Surge AI, Invisible, Turing, Mercor, TELUS Digital and Scale AI as AI-data competitors; Prolific work has many qualified homes. Date: EDGAR filing date."},{"url":"https://www.prolific.com/resources/prolific-in-2025-more-precision-larger-scale-and-stronger-safeguards-for-quality-human-data","date":"2025-12-16","support":"Prolific's HUMAINE study used nearly 27,000 evaluators (paragraph 7); Scale also runs demographic-segmented public evaluations. Full date."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02-26","support":"Innodata's 10-K (Item 1A) says customer agreements are generally terminable on about 30-90 days' notice, including on selecting another provider; Prolific volume can be moved on that timescale. Date: EDGAR filing date."},{"url":"https://fortune.com/2025/06/19/openai-is-phasing-out-scale-ai-work-following-startups-meta-deal/","date":"2025-06-19","support":"OpenAI phased out Scale AI work within days of the Meta deal and relied increasingly on other providers such as Mercor (paragraphs 1, 9); Prolific's expert-data work is similarly portable. Full date."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://techcrunch.com/2026/06/29/arena-the-ai-leaderboard-everyone-uses-is-now-a-100m-business/","date":"2026-06-29","support":"TechCrunch says human-labelling firms Mercor, Surge and Scale AI compete with Arena \"for the same dollar\" in evaluation (paragraph 7); Prolific's evaluation work has several qualified substitutes. Full date."}]}}}},"b":{"schema":"5lc-index-rating/1","key":"models-data-evaluation","entityId":"prolific","rater":"b","model":"claude-sonnet-5-5","ratedAt":"2026-10-22","functions":{"expert-data":{"alternatives":{"level":1,"evidence":[{"url":"https://sacra.com/research/350m-year-mercor-for-human-personality/","date":"2026-05-19","support":"Sacra names Mercor, Scale, Surge, Handshake, Turing, micro1 and Invisible as Prolific's competitors for frontier-lab work (19 May 2026; revenue figures are Sacra estimates)."},{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02","support":"Competition section names nine AI-data specialists (including Appen, CloudFactory, Surge, Invisible, Turing, Mercor, TELUS Digital and Scale) plus large services firms; Prolific is one of many. Date: month only, taken from the 20 Feb 2026 cover-page share count for the FY2025 filing."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://www.storyboard18.com/how-it-works/google-to-sever-ties-with-scale-ai-following-metas-significant-stake-70359.htm","date":"2025-06-16","support":"Google reportedly planned to end about $200M a year of Scale spend after Meta's stake; Handshake's CEO said demand 'tripled overnight' and Labelbox expected hundreds of millions in new revenue (report of 16 Jun 2025)."},{"url":"https://siliconangle.com/2025/07/01/data-labeling-startup-surge-ai-seeking-1b-first-capital-raise-reports-say/","date":"2025-07-01","support":"Surge reportedly won former Scale customers including OpenAI and Google by 1 July 2025, under three weeks after Meta's stake was confirmed on 13 June (SiliconANGLE, citing reports)."},{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02","support":"Innodata says its customer agreements are generally terminable on 30 to 90 days' notice and do not commit customers to future purchases (risk factors). Date: month only."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://techcrunch.com/2026/07/08/these-ai-startups-are-growing-revenue-at-faster-and-faster-rates/","date":"2026-07-08","support":"Mercor says (CEO, TechCrunch 8 Jul 2026) gross annualised revenue rose from $500M (Sept 2025) to $1bn and then over $2bn (June 2026); Handshake and micro1 report about $1bn each (company statements), so several vendors already operate at the billion-dollar scale."},{"url":"https://newsletter.semianalysis.com/p/rl-environments-and-rl-for-science","date":"2026-01-06","support":"Multiple labs 'mostly stopped contracting' Scale after Meta's stake and the work was absorbed by others such as Surge; the article names no capability shortfall in the replacements (6 Jan 2026)."}]}},"labelling-multimodal":{"alternatives":{"level":2,"evidence":[{"url":"https://sacra.com/research/350m-year-mercor-for-human-personality/","date":"2026-05-19","support":"Prolific sells ID-verified participants for demographic-targeted human data; Sacra says Mechanical Turk became gamed by bots, so I count only a few panels that fit that use (judgement; 19 May 2026; evidence thin)."},{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02","support":"Competition section names nine AI-data specialists (including Appen, CloudFactory, Surge, Invisible, Turing, Mercor, TELUS Digital and Scale) plus large services firms; Prolific is one of many. Date: month only, taken from the 20 Feb 2026 cover-page share count for the FY2025 filing."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://www.prolific.com/","date":"undated","support":"Ai2 says Prolific cut its human data collection 'from weeks to hours', so switching onto or off a panel is quick (page undated; the footer shows only a 2026 copyright)."},{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02","support":"Innodata says its customer agreements are generally terminable on 30 to 90 days' notice and do not commit customers to future purchases (risk factors). Date: month only."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://www.investegate.co.uk/announcement/rns/rws-holdings--rws/half-year-financial-report/9612324","date":"2026-06-11","support":"RWS TrainAI revenue (Generate segment) grew 44.7% to GBP 99.4M in the half to March 2026, with a new global technology client won, so incumbents are adding programmes and volume has somewhere to go (11 Jun 2026)."},{"url":"https://www.sec.gov/Archives/edgar/data/0000868675/000110465926088975/tm2615221d2_ex99-1.htm","date":"2026-07-31","support":"TELUS Digital's AI and data solutions line is shrinking on client ramp-downs, which suggests spare capacity among incumbents (31 Jul 2026)."}]}},"evaluation":{"alternatives":{"level":2,"evidence":[{"url":"https://www.prolific.com/","date":"undated","support":"Prolific sells human feedback from representative populations and runs the HUMAINE leaderboard; Toloka, Welo and Scale sell evaluation but without Prolific's verified-participant provenance (page undated; the footer shows only a 2026 copyright)."},{"url":"https://sacra.com/research/350m-year-mercor-for-human-personality/","date":"2026-05-19","support":"Prolific emphasises independence, trust record and deep participant pools versus bot-prone crowd platforms (19 May 2026)."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://www.prolific.com/","date":"undated","support":"Ai2 reports human data collection 'from weeks to hours' with Prolific (page undated; the footer shows only a 2026 copyright)."},{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02","support":"Innodata says its customer agreements are generally terminable on 30 to 90 days' notice and do not commit customers to future purchases (risk factors). Date: month only."}]},"capabilityGap":{"level":2,"evidence":[{"url":"https://sacra.com/research/350m-year-mercor-for-human-personality/","date":"2026-05-19","support":"Prolific began as a provenance-first marketplace with ID-verified participants; rivals such as Toloka and Welo offer evaluation but Sacra gives no evidence they match its verification (19 May 2026; thin evidence)."},{"url":"https://welodata.ai/","date":"undated","support":"Welo Data offers evaluation and red teaming with 500k+ curated experts (page undated; the footer shows only a 2026 copyright)."}]}}}}},"appen":{"a":{"schema":"5lc-index-rating/1","key":"models-data-evaluation","entityId":"appen","rater":"a","model":"claude-opus-5-5","ratedAt":"2026-10-22","functions":{"expert-data":{"alternatives":{"level":1,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02-26","support":"Innodata's FY2025 10-K (Item 1, Competition) names Appen, CloudFactory, Surge AI, Invisible, Turing, Mercor, TELUS Digital and Scale AI as AI-data competitors; Appen work has many qualified homes. Date: EDGAR filing date."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://fortune.com/2025/06/19/openai-is-phasing-out-scale-ai-work-following-startups-meta-deal/","date":"2025-06-19","support":"OpenAI phased out Scale AI work within days of the Meta deal and relied increasingly on other providers such as Mercor (paragraphs 1, 9); Appen's expert-data work is similarly portable. Full date."},{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02-26","support":"Innodata's 10-K (Item 1A) says customer agreements are generally terminable on about 30-90 days' notice, including on selecting another provider; Appen volume can be moved on that timescale. Date: EDGAR filing date."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://techcrunch.com/2026/09/22/snorkel-ai-triples-valuation-to-3-5b-as-demand-for-ai-training-data-booms/","date":"2026-09-22","support":"TechCrunch names Mercor (US$2bn gross), Handshake (US$1bn) and micro1 (US$500m) as scaled peers, and Snorkel at US$375m run-rate (company says) (paragraphs 4-5); several large alternatives to Appen. Full date."}]}},"labelling-multimodal":{"alternatives":{"level":1,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02-26","support":"Innodata's FY2025 10-K (Item 1, Competition) names Appen, CloudFactory, Surge AI, Invisible, Turing, Mercor, TELUS Digital and Scale AI as AI-data competitors; Appen work has many qualified homes. Date: EDGAR filing date."},{"url":"https://www.sec.gov/Archives/edgar/data/1829864/000182986426000012/task-20251231.htm","date":"2026-03-05","support":"TaskUs's FY2025 10-K (Item 1, Our Competition) calls AI data services a \"large, rapidly changing, and fragmented global market\"; many vendors can take Appen's labelling work. Date: EDGAR filing date."},{"url":"https://www.searchenginejournal.com/google-terminates-contract-with-appen-for-search-quality-raters/506006/","date":"2024-01-21","support":"Google gave Appen notice on 20 January 2024 and ended its services by 19 March 2024 (paragraphs 1-2), a two-month exit; Appen's volume could move similarly. Full date."}]},"switchingTime":{"level":2,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/1829864/000182986426000012/task-20251231.htm","date":"2026-03-05","support":"TaskUs's 10-K (Item 1A, labour risks) says hiring and training staff for increased demand \"takes time\"; absorbing Appen's large programmes would take months, not weeks. Date: EDGAR filing date."},{"url":"https://www.searchenginejournal.com/google-terminates-contract-with-appen-for-search-quality-raters/506006/","date":"2024-01-21","support":"Google gave Appen notice on 20 January 2024 and ended its services by 19 March 2024 (paragraphs 1-2), a two-month exit; Appen's volume could move similarly. Full date."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/1829864/000182986426000012/task-20251231.htm","date":"2026-03-05","support":"TaskUs's 10-K (Item 1) puts AI Services at 18% of US$1,183.5m 2025 revenue (about US$213m, 5LayerCake arithmetic), so rivals of Appen already run labelling at large scale. Date: EDGAR filing date."}]}}}},"b":{"schema":"5lc-index-rating/1","key":"models-data-evaluation","entityId":"appen","rater":"b","model":"claude-sonnet-5-5","ratedAt":"2026-10-22","functions":{"expert-data":{"alternatives":{"level":1,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02","support":"Competition section names nine AI-data specialists (including Appen, CloudFactory, Surge, Invisible, Turing, Mercor, TELUS Digital and Scale) plus large services firms; Appen is one of many. Date: month only, taken from the 20 Feb 2026 cover-page share count for the FY2025 filing."},{"url":"https://time.com/collection/time100-ai/2026/garrett-lord/","date":"2026-08-27","support":"Handshake AI says it works with eight frontier labs (27 Aug 2026), so several expert-data vendors are already qualified at labs."},{"url":"https://nebius.com/newsroom/nebius-welcomes-bezos-expeditions-as-lead-investor-in-ai-data-business-toloka","date":"2025-05-07","support":"Toloka lists Amazon, Anthropic, Microsoft, poolside, Recraft and Shopify as customers (7 May 2025), another vendor qualified at major buyers."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://www.storyboard18.com/how-it-works/google-to-sever-ties-with-scale-ai-following-metas-significant-stake-70359.htm","date":"2025-06-16","support":"Google reportedly planned to end about $200M a year of Scale spend after Meta's stake; Handshake's CEO said demand 'tripled overnight' and Labelbox expected hundreds of millions in new revenue (report of 16 Jun 2025)."},{"url":"https://siliconangle.com/2025/07/01/data-labeling-startup-surge-ai-seeking-1b-first-capital-raise-reports-say/","date":"2025-07-01","support":"Surge reportedly won former Scale customers including OpenAI and Google by 1 July 2025, under three weeks after Meta's stake was confirmed on 13 June (SiliconANGLE, citing reports)."},{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02","support":"Innodata says its customer agreements are generally terminable on 30 to 90 days' notice and do not commit customers to future purchases (risk factors). Date: month only."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://techcrunch.com/2026/07/08/these-ai-startups-are-growing-revenue-at-faster-and-faster-rates/","date":"2026-07-08","support":"Mercor says (CEO, TechCrunch 8 Jul 2026) gross annualised revenue rose from $500M (Sept 2025) to $1bn and then over $2bn (June 2026); Handshake and micro1 report about $1bn each (company statements), so several vendors already operate at the billion-dollar scale."},{"url":"https://newsletter.semianalysis.com/p/rl-environments-and-rl-for-science","date":"2026-01-06","support":"Multiple labs 'mostly stopped contracting' Scale after Meta's stake and the work was absorbed by others such as Surge; the article names no capability shortfall in the replacements (6 Jan 2026)."}]}},"labelling-multimodal":{"alternatives":{"level":1,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02","support":"Competition section names nine AI-data specialists (including Appen, CloudFactory, Surge, Invisible, Turing, Mercor, TELUS Digital and Scale) plus large services firms; Appen is one of many. Date: month only, taken from the 20 Feb 2026 cover-page share count for the FY2025 filing."},{"url":"https://www.sec.gov/Archives/edgar/data/0001829864/000182986426000147/earningsreleaseex991q22026.htm","date":"2026-08-05","support":"TaskUs reports AI Services as its fastest-growing line, up 25.8% year over year in Q2 2026, a large listed provider of AI data work (release of 5 Aug 2026)."},{"url":"https://announcements.asx.com.au/asxpdf/20260827/pdf/0739d4q7lyh3w2.pdf","date":"2026-08-27","support":"Appen China's annualised run-rate passed $175M in June 2026 and Appen Global is rebuilding; Appen is a further large vendor (ASX, 27 Aug 2026)."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://announcements.asx.com.au/asxpdf/20250226/pdf/06fyjdglv66466.pdf","date":"2025-02-26","support":"Google's termination of its Appen contract in Q1 FY24 cut Appen's FY24 group revenue 14.2% to $234.3M, showing a large buyer can withdraw a labelling programme within a quarter (ASX, 26 Feb 2025)."},{"url":"https://announcements.asx.com.au/asxpdf/20260827/pdf/0739d4q7lyh3w2.pdf","date":"2026-08-27","support":"Appen Global revenue fell 27% in H1 FY26 on lower volumes from traditional work while Appen China rose 80%, so volume can leave a vendor within a half-year (27 Aug 2026)."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://www.investegate.co.uk/announcement/rns/rws-holdings--rws/half-year-financial-report/9612324","date":"2026-06-11","support":"RWS TrainAI revenue (Generate segment) grew 44.7% to GBP 99.4M in the half to March 2026, with a new global technology client won, so incumbents are adding programmes and volume has somewhere to go (11 Jun 2026)."},{"url":"https://www.sec.gov/Archives/edgar/data/0000868675/000110465926088975/tm2615221d2_ex99-1.htm","date":"2026-07-31","support":"TELUS Digital's AI and data solutions line is shrinking on client ramp-downs, which suggests spare capacity among incumbents (31 Jul 2026)."}]}}}}},"taskus":{"a":{"schema":"5lc-index-rating/1","key":"models-data-evaluation","entityId":"taskus","rater":"a","model":"claude-opus-5-5","ratedAt":"2026-10-22","functions":{"expert-data":{"alternatives":{"level":1,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02-26","support":"Innodata's FY2025 10-K (Item 1, Competition) names Appen, CloudFactory, Surge AI, Invisible, Turing, Mercor, TELUS Digital and Scale AI as AI-data competitors; TaskUs work has many qualified homes. Date: EDGAR filing date."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://fortune.com/2025/06/19/openai-is-phasing-out-scale-ai-work-following-startups-meta-deal/","date":"2025-06-19","support":"OpenAI phased out Scale AI work within days of the Meta deal and relied increasingly on other providers such as Mercor (paragraphs 1, 9); TaskUs's expert-data work is similarly portable. Full date."},{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02-26","support":"Innodata's 10-K (Item 1A) says customer agreements are generally terminable on about 30-90 days' notice, including on selecting another provider; TaskUs volume can be moved on that timescale. Date: EDGAR filing date."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://techcrunch.com/2026/09/22/snorkel-ai-triples-valuation-to-3-5b-as-demand-for-ai-training-data-booms/","date":"2026-09-22","support":"TechCrunch names Mercor (US$2bn gross), Handshake (US$1bn) and micro1 (US$500m) as scaled peers, and Snorkel at US$375m run-rate (company says) (paragraphs 4-5); several large alternatives to TaskUs. Full date."}]}},"labelling-multimodal":{"alternatives":{"level":1,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02-26","support":"Innodata's FY2025 10-K (Item 1, Competition) names Appen, CloudFactory, Surge AI, Invisible, Turing, Mercor, TELUS Digital and Scale AI as AI-data competitors; TaskUs work has many qualified homes. Date: EDGAR filing date."},{"url":"https://www.sec.gov/Archives/edgar/data/1829864/000182986426000012/task-20251231.htm","date":"2026-03-05","support":"TaskUs's FY2025 10-K (Item 1, Our Competition) calls AI data services a \"large, rapidly changing, and fragmented global market\"; many vendors can take TaskUs's labelling work. Date: EDGAR filing date."}]},"switchingTime":{"level":2,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/1829864/000182986426000012/task-20251231.htm","date":"2026-03-05","support":"TaskUs's 10-K (Item 1A, labour risks) says hiring and training staff for increased demand \"takes time\"; absorbing TaskUs's large programmes would take months, not weeks. Date: EDGAR filing date."},{"url":"https://www.searchenginejournal.com/google-terminates-contract-with-appen-for-search-quality-raters/506006/","date":"2024-01-21","support":"Google gave Appen notice on 20 January 2024 and ended its services by 19 March 2024 (paragraphs 1-2), a two-month exit; TaskUs's volume could move similarly. Full date."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/1829864/000182986426000012/task-20251231.htm","date":"2026-03-05","support":"TaskUs's 10-K (Item 1) puts AI Services at 18% of US$1,183.5m 2025 revenue (about US$213m, 5LayerCake arithmetic), so rivals of TaskUs already run labelling at large scale. Date: EDGAR filing date."}]}},"evaluation":{"alternatives":{"level":1,"evidence":[{"url":"https://techcrunch.com/2026/06/29/arena-the-ai-leaderboard-everyone-uses-is-now-a-100m-business/","date":"2026-06-29","support":"TechCrunch says human-labelling firms Mercor, Surge and Scale AI compete with Arena \"for the same dollar\" in evaluation (paragraph 7); TaskUs's evaluation work has several qualified substitutes. Full date."},{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02-26","support":"Innodata's FY2025 10-K (Item 1, Competition) names Appen, CloudFactory, Surge AI, Invisible, Turing, Mercor, TELUS Digital and Scale AI as AI-data competitors; TaskUs work has many qualified homes. Date: EDGAR filing date."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02-26","support":"Innodata's 10-K (Item 1A) says customer agreements are generally terminable on about 30-90 days' notice, including on selecting another provider; TaskUs volume can be moved on that timescale. Date: EDGAR filing date."},{"url":"https://fortune.com/2025/06/19/openai-is-phasing-out-scale-ai-work-following-startups-meta-deal/","date":"2025-06-19","support":"OpenAI phased out Scale AI work within days of the Meta deal and relied increasingly on other providers such as Mercor (paragraphs 1, 9); TaskUs's expert-data work is similarly portable. Full date."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://techcrunch.com/2026/06/29/arena-the-ai-leaderboard-everyone-uses-is-now-a-100m-business/","date":"2026-06-29","support":"TechCrunch says human-labelling firms Mercor, Surge and Scale AI compete with Arena \"for the same dollar\" in evaluation (paragraph 7); TaskUs's evaluation work has several qualified substitutes. Full date."}]}}}},"b":{"schema":"5lc-index-rating/1","key":"models-data-evaluation","entityId":"taskus","rater":"b","model":"claude-sonnet-5-5","ratedAt":"2026-10-22","functions":{"expert-data":{"alternatives":{"level":1,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02","support":"Competition section names nine AI-data specialists (including Appen, CloudFactory, Surge, Invisible, Turing, Mercor, TELUS Digital and Scale) plus large services firms; TaskUs is one of many. Date: month only, taken from the 20 Feb 2026 cover-page share count for the FY2025 filing."},{"url":"https://time.com/collection/time100-ai/2026/garrett-lord/","date":"2026-08-27","support":"Handshake AI says it works with eight frontier labs (27 Aug 2026), so several expert-data vendors are already qualified at labs."},{"url":"https://nebius.com/newsroom/nebius-welcomes-bezos-expeditions-as-lead-investor-in-ai-data-business-toloka","date":"2025-05-07","support":"Toloka lists Amazon, Anthropic, Microsoft, poolside, Recraft and Shopify as customers (7 May 2025), another vendor qualified at major buyers."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://www.storyboard18.com/how-it-works/google-to-sever-ties-with-scale-ai-following-metas-significant-stake-70359.htm","date":"2025-06-16","support":"Google reportedly planned to end about $200M a year of Scale spend after Meta's stake; Handshake's CEO said demand 'tripled overnight' and Labelbox expected hundreds of millions in new revenue (report of 16 Jun 2025)."},{"url":"https://siliconangle.com/2025/07/01/data-labeling-startup-surge-ai-seeking-1b-first-capital-raise-reports-say/","date":"2025-07-01","support":"Surge reportedly won former Scale customers including OpenAI and Google by 1 July 2025, under three weeks after Meta's stake was confirmed on 13 June (SiliconANGLE, citing reports)."},{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02","support":"Innodata says its customer agreements are generally terminable on 30 to 90 days' notice and do not commit customers to future purchases (risk factors). Date: month only."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://techcrunch.com/2026/07/08/these-ai-startups-are-growing-revenue-at-faster-and-faster-rates/","date":"2026-07-08","support":"Mercor says (CEO, TechCrunch 8 Jul 2026) gross annualised revenue rose from $500M (Sept 2025) to $1bn and then over $2bn (June 2026); Handshake and micro1 report about $1bn each (company statements), so several vendors already operate at the billion-dollar scale."},{"url":"https://newsletter.semianalysis.com/p/rl-environments-and-rl-for-science","date":"2026-01-06","support":"Multiple labs 'mostly stopped contracting' Scale after Meta's stake and the work was absorbed by others such as Surge; the article names no capability shortfall in the replacements (6 Jan 2026)."}]}},"labelling-multimodal":{"alternatives":{"level":1,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02","support":"Competition section names nine AI-data specialists (including Appen, CloudFactory, Surge, Invisible, Turing, Mercor, TELUS Digital and Scale) plus large services firms; TaskUs is one of many. Date: month only, taken from the 20 Feb 2026 cover-page share count for the FY2025 filing."},{"url":"https://www.sec.gov/Archives/edgar/data/0001829864/000182986426000147/earningsreleaseex991q22026.htm","date":"2026-08-05","support":"TaskUs reports AI Services as its fastest-growing line, up 25.8% year over year in Q2 2026, a large listed provider of AI data work (release of 5 Aug 2026)."},{"url":"https://announcements.asx.com.au/asxpdf/20260827/pdf/0739d4q7lyh3w2.pdf","date":"2026-08-27","support":"Appen China's annualised run-rate passed $175M in June 2026 and Appen Global is rebuilding; Appen is a further large vendor (ASX, 27 Aug 2026)."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://announcements.asx.com.au/asxpdf/20250226/pdf/06fyjdglv66466.pdf","date":"2025-02-26","support":"Google's termination of its Appen contract in Q1 FY24 cut Appen's FY24 group revenue 14.2% to $234.3M, showing a large buyer can withdraw a labelling programme within a quarter (ASX, 26 Feb 2025)."},{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02","support":"Innodata says its customer agreements are generally terminable on 30 to 90 days' notice and do not commit customers to future purchases (risk factors). Date: month only."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://www.investegate.co.uk/announcement/rns/rws-holdings--rws/half-year-financial-report/9612324","date":"2026-06-11","support":"RWS TrainAI revenue (Generate segment) grew 44.7% to GBP 99.4M in the half to March 2026, with a new global technology client won, so incumbents are adding programmes and volume has somewhere to go (11 Jun 2026)."},{"url":"https://www.sec.gov/Archives/edgar/data/0000868675/000110465926088975/tm2615221d2_ex99-1.htm","date":"2026-07-31","support":"TELUS Digital's AI and data solutions line is shrinking on client ramp-downs, which suggests spare capacity among incumbents (31 Jul 2026)."}]}},"evaluation":{"alternatives":{"level":1,"evidence":[{"url":"https://labs.scale.com/leaderboard","date":"undated","support":"Scale Labs publishes 19 active leaderboards (page undated; the footer shows only a 2026 copyright)."},{"url":"https://www.prolific.com/","date":"undated","support":"Prolific sells human feedback for preference tuning, safety evals and benchmarks and lists a HUMAINE leaderboard (page undated; the footer shows only a 2026 copyright)."},{"url":"https://nebius.com/newsroom/nebius-welcomes-bezos-expeditions-as-lead-investor-in-ai-data-business-toloka","date":"2025-05-07","support":"Toloka offers agent safety evaluation and red teaming to labs (Nebius announcement, 7 May 2025)."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://mercor.com/blog/introducing-apex-ai-productivity-index/","date":"2025-10-01","support":"Mercor built its APEX v1 test set with about 100 experts in four professions (200 of 300 prompts approved), suggesting a comparable expert-written set is a matter of months (my inference, 1 Oct 2025)."},{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02","support":"Innodata says its customer agreements are generally terminable on 30 to 90 days' notice and do not commit customers to future purchases (risk factors). Date: month only."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://scale.com/blog/scales-next-era-building-for-2026","date":"2026-01-22","support":"Scale's lab says its lab introduced 15 new benchmarks and published over 450 evaluations of more than 50 models, a scale other vendors also reach (company blog, 22 Jan 2026)."},{"url":"https://welodata.ai/","date":"undated","support":"Welo Data lists evaluations, benchmarking, model selection and red teaming among its services (page undated; the footer shows only a 2026 copyright)."}]}}}}},"invisible-technologies":{"a":{"schema":"5lc-index-rating/1","key":"models-data-evaluation","entityId":"invisible-technologies","rater":"a","model":"claude-opus-5-5","ratedAt":"2026-10-22","functions":{"expert-data":{"alternatives":{"level":1,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02-26","support":"Innodata's FY2025 10-K (Item 1, Competition) names Appen, CloudFactory, Surge AI, Invisible, Turing, Mercor, TELUS Digital and Scale AI as AI-data competitors; Invisible work has many qualified homes. Date: EDGAR filing date."},{"url":"https://siliconangle.com/2025/09/16/ai-data-provider-invisible-raises-100m-2b-valuation/","date":"2025-09-16","support":"SiliconANGLE describes Invisible's RLHF and expert-annotation work and names Surge AI as a competitor (paragraphs 5, 10). Full date."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://fortune.com/2025/06/19/openai-is-phasing-out-scale-ai-work-following-startups-meta-deal/","date":"2025-06-19","support":"OpenAI phased out Scale AI work within days of the Meta deal and relied increasingly on other providers such as Mercor (paragraphs 1, 9); Invisible's expert-data work is similarly portable. Full date."},{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02-26","support":"Innodata's 10-K (Item 1A) says customer agreements are generally terminable on about 30-90 days' notice, including on selecting another provider; Invisible volume can be moved on that timescale. Date: EDGAR filing date."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://techcrunch.com/2026/09/22/snorkel-ai-triples-valuation-to-3-5b-as-demand-for-ai-training-data-booms/","date":"2026-09-22","support":"TechCrunch names Mercor (US$2bn gross), Handshake (US$1bn) and micro1 (US$500m) as scaled peers, and Snorkel at US$375m run-rate (company says) (paragraphs 4-5); several large alternatives to Invisible. Full date."}]}},"labelling-multimodal":{"alternatives":{"level":1,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02-26","support":"Innodata's FY2025 10-K (Item 1, Competition) names Appen, CloudFactory, Surge AI, Invisible, Turing, Mercor, TELUS Digital and Scale AI as AI-data competitors; Invisible work has many qualified homes. Date: EDGAR filing date."},{"url":"https://www.sec.gov/Archives/edgar/data/1829864/000182986426000012/task-20251231.htm","date":"2026-03-05","support":"TaskUs's FY2025 10-K (Item 1, Our Competition) calls AI data services a \"large, rapidly changing, and fragmented global market\"; many vendors can take Invisible's labelling work. Date: EDGAR filing date."},{"url":"https://siliconangle.com/2025/09/16/ai-data-provider-invisible-raises-100m-2b-valuation/","date":"2025-09-16","support":"Invisible's 2024 revenue was US$134m (paragraph 9); its labelling work is one among many vendors'. Full date."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://www.searchenginejournal.com/google-terminates-contract-with-appen-for-search-quality-raters/506006/","date":"2024-01-21","support":"Google gave Appen notice on 20 January 2024 and ended its services by 19 March 2024 (paragraphs 1-2), a two-month exit; Invisible's volume could move similarly. Full date."},{"url":"https://www.sec.gov/Archives/edgar/data/1829864/000182986426000012/task-20251231.htm","date":"2026-03-05","support":"TaskUs's 10-K (Item 1A) says clients can terminate for convenience or move work in-house or to competitors; Invisible's programmes are portable within weeks to months. Date: EDGAR filing date."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/1829864/000182986426000012/task-20251231.htm","date":"2026-03-05","support":"TaskUs's 10-K (Item 1) puts AI Services at 18% of US$1,183.5m 2025 revenue (about US$213m, 5LayerCake arithmetic), so rivals of Invisible already run labelling at large scale. Date: EDGAR filing date."}]}},"rl-environments":{"alternatives":{"level":1,"evidence":[{"url":"https://newsletter.semianalysis.com/p/rl-environments-and-rl-for-science","date":"2026-01-06","support":"SemiAnalysis counts \"more than 35 companies\" building RL environments, mostly seed-stage, each serving 1-3 customers (vendor section); Invisible faces many substitutes. Full date."},{"url":"https://epoch.ai/gradient-updates/state-of-rl-envs","date":"2026-01-12","support":"Epoch's RL-environment FAQ lists specialised startups, data providers (Mercor, Surge, Handshake, Turing), product companies and labs' in-house teams as sources ('Which companies build RL Environments?'); many alternatives to Invisible. Full date."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://techcrunch.com/2025/09/16/silicon-valley-bets-big-on-environments-to-train-ai-agents/","date":"2025-09-16","support":"TechCrunch says \"there's no shortage of startups\" supplying environments and that big labs also build them in-house (paragraphs 3-4); alternatives to Invisible are plentiful. Full date."},{"url":"https://newsletter.semianalysis.com/p/rl-environments-and-rl-for-science","date":"2026-01-06","support":"SemiAnalysis says most environment vendors sell under exclusive contracts and UI-gym sites cost about US$20,000 each as one-time purchases (vendor and pricing sections); replacing Invisible's pipeline means commissioning new builds. Full date."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://techcrunch.com/2026/09/22/snorkel-ai-triples-valuation-to-3-5b-as-demand-for-ai-training-data-booms/","date":"2026-09-22","support":"TechCrunch names Mercor (US$2bn gross), Handshake (US$1bn) and micro1 (US$500m) as scaled peers, and Snorkel at US$375m run-rate (company says) (paragraphs 4-5); several large alternatives to Invisible. Full date."}]}}}},"b":{"schema":"5lc-index-rating/1","key":"models-data-evaluation","entityId":"invisible-technologies","rater":"b","model":"claude-sonnet-5-5","ratedAt":"2026-10-22","functions":{"expert-data":{"alternatives":{"level":1,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02","support":"Competition section names nine AI-data specialists (including Appen, CloudFactory, Surge, Invisible, Turing, Mercor, TELUS Digital and Scale) plus large services firms; Invisible is one of many. Date: month only, taken from the 20 Feb 2026 cover-page share count for the FY2025 filing."},{"url":"https://time.com/collection/time100-ai/2026/garrett-lord/","date":"2026-08-27","support":"Handshake AI says it works with eight frontier labs (27 Aug 2026), so several expert-data vendors are already qualified at labs."},{"url":"https://nebius.com/newsroom/nebius-welcomes-bezos-expeditions-as-lead-investor-in-ai-data-business-toloka","date":"2025-05-07","support":"Toloka lists Amazon, Anthropic, Microsoft, poolside, Recraft and Shopify as customers (7 May 2025), another vendor qualified at major buyers."},{"url":"https://siliconangle.com/2025/09/16/ai-data-provider-invisible-raises-100m-2b-valuation/","date":"2025-09-16","support":"SiliconANGLE names Surge AI as a rival of Invisible, which reported $134M of revenue in 2024 (16 Sep 2025)."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://www.storyboard18.com/how-it-works/google-to-sever-ties-with-scale-ai-following-metas-significant-stake-70359.htm","date":"2025-06-16","support":"Google reportedly planned to end about $200M a year of Scale spend after Meta's stake; Handshake's CEO said demand 'tripled overnight' and Labelbox expected hundreds of millions in new revenue (report of 16 Jun 2025)."},{"url":"https://siliconangle.com/2025/07/01/data-labeling-startup-surge-ai-seeking-1b-first-capital-raise-reports-say/","date":"2025-07-01","support":"Surge reportedly won former Scale customers including OpenAI and Google by 1 July 2025, under three weeks after Meta's stake was confirmed on 13 June (SiliconANGLE, citing reports)."},{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02","support":"Innodata says its customer agreements are generally terminable on 30 to 90 days' notice and do not commit customers to future purchases (risk factors). Date: month only."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://techcrunch.com/2026/07/08/these-ai-startups-are-growing-revenue-at-faster-and-faster-rates/","date":"2026-07-08","support":"Mercor says (CEO, TechCrunch 8 Jul 2026) gross annualised revenue rose from $500M (Sept 2025) to $1bn and then over $2bn (June 2026); Handshake and micro1 report about $1bn each (company statements), so several vendors already operate at the billion-dollar scale."},{"url":"https://newsletter.semianalysis.com/p/rl-environments-and-rl-for-science","date":"2026-01-06","support":"Multiple labs 'mostly stopped contracting' Scale after Meta's stake and the work was absorbed by others such as Surge; the article names no capability shortfall in the replacements (6 Jan 2026)."}]}},"labelling-multimodal":{"alternatives":{"level":1,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02","support":"Competition section names nine AI-data specialists (including Appen, CloudFactory, Surge, Invisible, Turing, Mercor, TELUS Digital and Scale) plus large services firms; Invisible is one of many. Date: month only, taken from the 20 Feb 2026 cover-page share count for the FY2025 filing."},{"url":"https://www.sec.gov/Archives/edgar/data/0001829864/000182986426000147/earningsreleaseex991q22026.htm","date":"2026-08-05","support":"TaskUs reports AI Services as its fastest-growing line, up 25.8% year over year in Q2 2026, a large listed provider of AI data work (release of 5 Aug 2026)."},{"url":"https://announcements.asx.com.au/asxpdf/20260827/pdf/0739d4q7lyh3w2.pdf","date":"2026-08-27","support":"Appen China's annualised run-rate passed $175M in June 2026 and Appen Global is rebuilding; Appen is a further large vendor (ASX, 27 Aug 2026)."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://announcements.asx.com.au/asxpdf/20250226/pdf/06fyjdglv66466.pdf","date":"2025-02-26","support":"Google's termination of its Appen contract in Q1 FY24 cut Appen's FY24 group revenue 14.2% to $234.3M, showing a large buyer can withdraw a labelling programme within a quarter (ASX, 26 Feb 2025)."},{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02","support":"Innodata says its customer agreements are generally terminable on 30 to 90 days' notice and do not commit customers to future purchases (risk factors). Date: month only."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://www.investegate.co.uk/announcement/rns/rws-holdings--rws/half-year-financial-report/9612324","date":"2026-06-11","support":"RWS TrainAI revenue (Generate segment) grew 44.7% to GBP 99.4M in the half to March 2026, with a new global technology client won, so incumbents are adding programmes and volume has somewhere to go (11 Jun 2026)."},{"url":"https://www.sec.gov/Archives/edgar/data/0000868675/000110465926088975/tm2615221d2_ex99-1.htm","date":"2026-07-31","support":"TELUS Digital's AI and data solutions line is shrinking on client ramp-downs, which suggests spare capacity among incumbents (31 Jul 2026)."}]}},"rl-environments":{"alternatives":{"level":1,"evidence":[{"url":"https://newsletter.semianalysis.com/p/rl-environments-and-rl-for-science","date":"2026-01-06","support":"More than 35 companies sell RL environments (mostly seed-stage) and Anthropic likely works with more than a dozen of them (SemiAnalysis, 6 Jan 2026)."},{"url":"https://epoch.ai/gradient-updates/state-of-rl-envs","date":"2026-01-12","support":"Epoch lists specialist startups, the data vendors Mercor, Surge, Handshake and Turing, in-house lab teams and product-company partners as sources of environments (12 Jan 2026)."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://epoch.ai/gradient-updates/state-of-rl-envs","date":"2026-01-12","support":"Environment contracts are often six to seven figures per quarter and can be non-exclusive, so labs can add or swap vendors quarter by quarter; Epoch also reports 'substantially more' in-housing (12 Jan 2026)."},{"url":"https://newsletter.semianalysis.com/p/rl-environments-and-rl-for-science","date":"2026-01-06","support":"Anthropic is 'often the first customer' of new environment startups and wants a broad vendor ecosystem so the product is commoditised (SemiAnalysis, 6 Jan 2026)."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://epoch.ai/gradient-updates/state-of-rl-envs","date":"2026-01-12","support":"Quality at scale is the main bottleneck for all vendors; interviewees rank robustness to reward hacking first and say domain knowledge matters more than ML skill (12 Jan 2026). No vendor is described as uniquely capable."},{"url":"https://newsletter.semianalysis.com/p/rl-environments-and-rl-for-science","date":"2026-01-06","support":"Reports 'significant diversity in quality or focus' among vendors, with Turing on UI gyms and others on software engineering, rather than a single leader (6 Jan 2026)."}]}}}}},"telus-digital":{"a":{"schema":"5lc-index-rating/1","key":"models-data-evaluation","entityId":"telus-digital","rater":"a","model":"claude-opus-5-5","ratedAt":"2026-10-22","functions":{"expert-data":{"alternatives":{"level":1,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02-26","support":"Innodata's FY2025 10-K (Item 1, Competition) names Appen, CloudFactory, Surge AI, Invisible, Turing, Mercor, TELUS Digital and Scale AI as AI-data competitors; TELUS Digital work has many qualified homes. Date: EDGAR filing date."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://fortune.com/2025/06/19/openai-is-phasing-out-scale-ai-work-following-startups-meta-deal/","date":"2025-06-19","support":"OpenAI phased out Scale AI work within days of the Meta deal and relied increasingly on other providers such as Mercor (paragraphs 1, 9); TELUS Digital's expert-data work is similarly portable. Full date."},{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02-26","support":"Innodata's 10-K (Item 1A) says customer agreements are generally terminable on about 30-90 days' notice, including on selecting another provider; TELUS Digital volume can be moved on that timescale. Date: EDGAR filing date."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://techcrunch.com/2026/09/22/snorkel-ai-triples-valuation-to-3-5b-as-demand-for-ai-training-data-booms/","date":"2026-09-22","support":"TechCrunch names Mercor (US$2bn gross), Handshake (US$1bn) and micro1 (US$500m) as scaled peers, and Snorkel at US$375m run-rate (company says) (paragraphs 4-5); several large alternatives to TELUS Digital. Full date."}]}},"labelling-multimodal":{"alternatives":{"level":1,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02-26","support":"Innodata's FY2025 10-K (Item 1, Competition) names Appen, CloudFactory, Surge AI, Invisible, Turing, Mercor, TELUS Digital and Scale AI as AI-data competitors; TELUS Digital work has many qualified homes. Date: EDGAR filing date."},{"url":"https://www.sec.gov/Archives/edgar/data/1829864/000182986426000012/task-20251231.htm","date":"2026-03-05","support":"TaskUs's FY2025 10-K (Item 1, Our Competition) calls AI data services a \"large, rapidly changing, and fragmented global market\"; many vendors can take TELUS Digital's labelling work. Date: EDGAR filing date."}]},"switchingTime":{"level":2,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/1829864/000182986426000012/task-20251231.htm","date":"2026-03-05","support":"TaskUs's 10-K (Item 1A, labour risks) says hiring and training staff for increased demand \"takes time\"; absorbing TELUS Digital's large programmes would take months, not weeks. Date: EDGAR filing date."},{"url":"https://www.searchenginejournal.com/google-terminates-contract-with-appen-for-search-quality-raters/506006/","date":"2024-01-21","support":"Google gave Appen notice on 20 January 2024 and ended its services by 19 March 2024 (paragraphs 1-2), a two-month exit; TELUS Digital's volume could move similarly. Full date."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/1829864/000182986426000012/task-20251231.htm","date":"2026-03-05","support":"TaskUs's 10-K (Item 1) puts AI Services at 18% of US$1,183.5m 2025 revenue (about US$213m, 5LayerCake arithmetic), so rivals of TELUS Digital already run labelling at large scale. Date: EDGAR filing date."}]}}}},"b":{"schema":"5lc-index-rating/1","key":"models-data-evaluation","entityId":"telus-digital","rater":"b","model":"claude-sonnet-5-5","ratedAt":"2026-10-22","functions":{"expert-data":{"alternatives":{"level":1,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02","support":"Competition section names nine AI-data specialists (including Appen, CloudFactory, Surge, Invisible, Turing, Mercor, TELUS Digital and Scale) plus large services firms; TELUS Digital is one of many. Date: month only, taken from the 20 Feb 2026 cover-page share count for the FY2025 filing."},{"url":"https://time.com/collection/time100-ai/2026/garrett-lord/","date":"2026-08-27","support":"Handshake AI says it works with eight frontier labs (27 Aug 2026), so several expert-data vendors are already qualified at labs."},{"url":"https://nebius.com/newsroom/nebius-welcomes-bezos-expeditions-as-lead-investor-in-ai-data-business-toloka","date":"2025-05-07","support":"Toloka lists Amazon, Anthropic, Microsoft, poolside, Recraft and Shopify as customers (7 May 2025), another vendor qualified at major buyers."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/0000868675/000110465926088975/tm2615221d2_ex99-1.htm","date":"2026-07-31","support":"TELUS reports 'client ramp-downs in our AI and data solutions service line' among the reasons TELUS Digital revenue fell 10% in Q2 2026 (release of 31 Jul 2026)."},{"url":"https://www.storyboard18.com/how-it-works/google-to-sever-ties-with-scale-ai-following-metas-significant-stake-70359.htm","date":"2025-06-16","support":"Google reportedly planned to end about $200M a year of Scale spend after Meta's stake; Handshake's CEO said demand 'tripled overnight' and Labelbox expected hundreds of millions in new revenue (report of 16 Jun 2025)."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://techcrunch.com/2026/07/08/these-ai-startups-are-growing-revenue-at-faster-and-faster-rates/","date":"2026-07-08","support":"Mercor says (CEO, TechCrunch 8 Jul 2026) gross annualised revenue rose from $500M (Sept 2025) to $1bn and then over $2bn (June 2026); Handshake and micro1 report about $1bn each (company statements), so several vendors already operate at the billion-dollar scale."},{"url":"https://newsletter.semianalysis.com/p/rl-environments-and-rl-for-science","date":"2026-01-06","support":"Multiple labs 'mostly stopped contracting' Scale after Meta's stake and the work was absorbed by others such as Surge; the article names no capability shortfall in the replacements (6 Jan 2026)."}]}},"labelling-multimodal":{"alternatives":{"level":1,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02","support":"Competition section names nine AI-data specialists (including Appen, CloudFactory, Surge, Invisible, Turing, Mercor, TELUS Digital and Scale) plus large services firms; TELUS Digital is one of many. Date: month only, taken from the 20 Feb 2026 cover-page share count for the FY2025 filing."},{"url":"https://www.sec.gov/Archives/edgar/data/0001829864/000182986426000147/earningsreleaseex991q22026.htm","date":"2026-08-05","support":"TaskUs reports AI Services as its fastest-growing line, up 25.8% year over year in Q2 2026, a large listed provider of AI data work (release of 5 Aug 2026)."},{"url":"https://announcements.asx.com.au/asxpdf/20260827/pdf/0739d4q7lyh3w2.pdf","date":"2026-08-27","support":"Appen China's annualised run-rate passed $175M in June 2026 and Appen Global is rebuilding; Appen is a further large vendor (ASX, 27 Aug 2026)."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/0000868675/000110465926088975/tm2615221d2_ex99-1.htm","date":"2026-07-31","support":"TELUS reports 'client ramp-downs in our AI and data solutions service line' among the reasons TELUS Digital revenue fell 10% in Q2 2026 (release of 31 Jul 2026)."},{"url":"https://announcements.asx.com.au/asxpdf/20250226/pdf/06fyjdglv66466.pdf","date":"2025-02-26","support":"Google's termination of its Appen contract in Q1 FY24 cut Appen's FY24 group revenue 14.2% to $234.3M, showing a large buyer can withdraw a labelling programme within a quarter (ASX, 26 Feb 2025)."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://www.investegate.co.uk/announcement/rns/rws-holdings--rws/half-year-financial-report/9612324","date":"2026-06-11","support":"RWS TrainAI revenue (Generate segment) grew 44.7% to GBP 99.4M in the half to March 2026, with a new global technology client won, so incumbents are adding programmes and volume has somewhere to go (11 Jun 2026)."},{"url":"https://www.sec.gov/Archives/edgar/data/0000868675/000110465926088975/tm2615221d2_ex99-1.htm","date":"2026-07-31","support":"TELUS Digital's AI and data solutions line is shrinking on client ramp-downs, which suggests spare capacity among incumbents (31 Jul 2026)."}]}}}}},"rws-trainai":{"a":{"schema":"5lc-index-rating/1","key":"models-data-evaluation","entityId":"rws-trainai","rater":"a","model":"claude-opus-5-5","ratedAt":"2026-10-22","functions":{"labelling-multimodal":{"alternatives":{"level":1,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02-26","support":"Innodata's FY2025 10-K (Item 1, Competition) names Appen, CloudFactory, Surge AI, Invisible, Turing, Mercor, TELUS Digital and Scale AI as AI-data competitors; RWS TrainAI work has many qualified homes. Date: EDGAR filing date."},{"url":"https://www.sec.gov/Archives/edgar/data/1829864/000182986426000012/task-20251231.htm","date":"2026-03-05","support":"TaskUs's FY2025 10-K (Item 1, Our Competition) calls AI data services a \"large, rapidly changing, and fragmented global market\"; many vendors can take RWS TrainAI's labelling work. Date: EDGAR filing date."}]},"switchingTime":{"level":2,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/1829864/000182986426000012/task-20251231.htm","date":"2026-03-05","support":"TaskUs's 10-K (Item 1A, labour risks) says hiring and training staff for increased demand \"takes time\"; absorbing RWS TrainAI's large programmes would take months, not weeks. Date: EDGAR filing date."},{"url":"https://www.searchenginejournal.com/google-terminates-contract-with-appen-for-search-quality-raters/506006/","date":"2024-01-21","support":"Google gave Appen notice on 20 January 2024 and ended its services by 19 March 2024 (paragraphs 1-2), a two-month exit; RWS TrainAI's volume could move similarly. Full date."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/1829864/000182986426000012/task-20251231.htm","date":"2026-03-05","support":"TaskUs's 10-K (Item 1) puts AI Services at 18% of US$1,183.5m 2025 revenue (about US$213m, 5LayerCake arithmetic), so rivals of RWS TrainAI already run labelling at large scale. Date: EDGAR filing date."}]}}}},"b":{"schema":"5lc-index-rating/1","key":"models-data-evaluation","entityId":"rws-trainai","rater":"b","model":"claude-sonnet-5-5","ratedAt":"2026-10-22","functions":{"labelling-multimodal":{"alternatives":{"level":1,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02","support":"Competition section names nine AI-data specialists (including Appen, CloudFactory, Surge, Invisible, Turing, Mercor, TELUS Digital and Scale) plus large services firms; RWS TrainAI is one of many. Date: month only, taken from the 20 Feb 2026 cover-page share count for the FY2025 filing."},{"url":"https://www.sec.gov/Archives/edgar/data/0001829864/000182986426000147/earningsreleaseex991q22026.htm","date":"2026-08-05","support":"TaskUs reports AI Services as its fastest-growing line, up 25.8% year over year in Q2 2026, a large listed provider of AI data work (release of 5 Aug 2026)."},{"url":"https://announcements.asx.com.au/asxpdf/20260827/pdf/0739d4q7lyh3w2.pdf","date":"2026-08-27","support":"Appen China's annualised run-rate passed $175M in June 2026 and Appen Global is rebuilding; Appen is a further large vendor (ASX, 27 Aug 2026)."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://announcements.asx.com.au/asxpdf/20250226/pdf/06fyjdglv66466.pdf","date":"2025-02-26","support":"Google's termination of its Appen contract in Q1 FY24 cut Appen's FY24 group revenue 14.2% to $234.3M, showing a large buyer can withdraw a labelling programme within a quarter (ASX, 26 Feb 2025)."},{"url":"https://www.investegate.co.uk/announcement/rns/rws-holdings--rws/half-year-financial-report/9612324","date":"2026-06-11","support":"RWS credits growth to an extra programme with an existing client and a newly won client, i.e. volume moves in programme-sized steps (11 Jun 2026)."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://www.investegate.co.uk/announcement/rns/rws-holdings--rws/half-year-financial-report/9612324","date":"2026-06-11","support":"RWS TrainAI revenue (Generate segment) grew 44.7% to GBP 99.4M in the half to March 2026, with a new global technology client won, so incumbents are adding programmes and volume has somewhere to go (11 Jun 2026)."},{"url":"https://www.sec.gov/Archives/edgar/data/0000868675/000110465926088975/tm2615221d2_ex99-1.htm","date":"2026-07-31","support":"TELUS Digital's AI and data solutions line is shrinking on client ramp-downs, which suggests spare capacity among incumbents (31 Jul 2026)."}]}}}}},"proximal":{"a":{"schema":"5lc-index-rating/1","key":"models-data-evaluation","entityId":"proximal","rater":"a","model":"claude-opus-5-5","ratedAt":"2026-10-22","functions":{"rl-environments":{"alternatives":{"level":1,"evidence":[{"url":"https://newsletter.semianalysis.com/p/rl-environments-and-rl-for-science","date":"2026-01-06","support":"SemiAnalysis counts \"more than 35 companies\" building RL environments, mostly seed-stage, each serving 1-3 customers (vendor section); Proximal faces many substitutes. Full date."},{"url":"https://techcrunch.com/2025/09/16/silicon-valley-bets-big-on-environments-to-train-ai-agents/","date":"2025-09-16","support":"TechCrunch says \"there's no shortage of startups\" supplying environments and that big labs also build them in-house (paragraphs 3-4); alternatives to Proximal are plentiful. Full date."}]},"switchingTime":{"level":2,"evidence":[{"url":"https://newsletter.semianalysis.com/p/rl-environments-and-rl-for-science","date":"2026-01-06","support":"SemiAnalysis says most environment vendors sell under exclusive contracts and UI-gym sites cost about US$20,000 each as one-time purchases (vendor and pricing sections); replacing Proximal's pipeline means commissioning new builds. Full date."},{"url":"https://techcrunch.com/2025/09/16/silicon-valley-bets-big-on-environments-to-train-ai-agents/","date":"2025-09-16","support":"Ross Taylor warns \"people are underestimating how difficult it is to scale environments\" (paragraph 32); a lab replacing Proximal's environment output would need months to re-commission it. Full date."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://epoch.ai/gradient-updates/state-of-rl-envs","date":"2026-01-12","support":"Epoch's RL-environment FAQ lists specialised startups, data providers (Mercor, Surge, Handshake, Turing), product companies and labs' in-house teams as sources ('Which companies build RL Environments?'); many alternatives to Proximal. Full date."}]}}}},"b":{"schema":"5lc-index-rating/1","key":"models-data-evaluation","entityId":"proximal","rater":"b","model":"claude-sonnet-5-5","ratedAt":"2026-10-22","functions":{"rl-environments":{"alternatives":{"level":1,"evidence":[{"url":"https://newsletter.semianalysis.com/p/rl-environments-and-rl-for-science","date":"2026-01-06","support":"More than 35 companies sell RL environments (mostly seed-stage) and Anthropic likely works with more than a dozen of them (SemiAnalysis, 6 Jan 2026)."},{"url":"https://startupfortune.com/proximal-reveals-200-million-revenue-run-rate-and-15-million-seed-round/","date":"2026-10-09","support":"Proximal says (reported 9 Oct 2026, self-reported, unverified) it reached $200M annualised revenue in under a year; the article names Standard Machines and Anthropic as others hiring for RL environments."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://epoch.ai/gradient-updates/state-of-rl-envs","date":"2026-01-12","support":"Environment contracts are often six to seven figures per quarter and can be non-exclusive, so labs can add or swap vendors quarter by quarter; Epoch also reports 'substantially more' in-housing (12 Jan 2026)."},{"url":"https://newsletter.semianalysis.com/p/rl-environments-and-rl-for-science","date":"2026-01-06","support":"Anthropic is 'often the first customer' of new environment startups and wants a broad vendor ecosystem so the product is commoditised (SemiAnalysis, 6 Jan 2026)."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://epoch.ai/gradient-updates/state-of-rl-envs","date":"2026-01-12","support":"Quality at scale is the main bottleneck for all vendors; interviewees rank robustness to reward hacking first and say domain knowledge matters more than ML skill (12 Jan 2026). No vendor is described as uniquely capable."},{"url":"https://newsletter.semianalysis.com/p/rl-environments-and-rl-for-science","date":"2026-01-06","support":"Reports 'significant diversity in quality or focus' among vendors, with Turing on UI gyms and others on software engineering, rather than a single leader (6 Jan 2026)."}]}}}}},"labelbox":{"a":{"schema":"5lc-index-rating/1","key":"models-data-evaluation","entityId":"labelbox","rater":"a","model":"claude-opus-5-5","ratedAt":"2026-10-22","functions":{"expert-data":{"alternatives":{"level":1,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02-26","support":"Innodata's FY2025 10-K (Item 1, Competition) names Appen, CloudFactory, Surge AI, Invisible, Turing, Mercor, TELUS Digital and Scale AI as AI-data competitors; Labelbox work has many qualified homes. Date: EDGAR filing date."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://fortune.com/2025/06/19/openai-is-phasing-out-scale-ai-work-following-startups-meta-deal/","date":"2025-06-19","support":"OpenAI phased out Scale AI work within days of the Meta deal and relied increasingly on other providers such as Mercor (paragraphs 1, 9); Labelbox's expert-data work is similarly portable. Full date."},{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02-26","support":"Innodata's 10-K (Item 1A) says customer agreements are generally terminable on about 30-90 days' notice, including on selecting another provider; Labelbox volume can be moved on that timescale. Date: EDGAR filing date."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://techcrunch.com/2026/09/22/snorkel-ai-triples-valuation-to-3-5b-as-demand-for-ai-training-data-booms/","date":"2026-09-22","support":"TechCrunch names Mercor (US$2bn gross), Handshake (US$1bn) and micro1 (US$500m) as scaled peers, and Snorkel at US$375m run-rate (company says) (paragraphs 4-5); several large alternatives to Labelbox. Full date."}]}},"evaluation":{"alternatives":{"level":1,"evidence":[{"url":"https://techcrunch.com/2026/06/29/arena-the-ai-leaderboard-everyone-uses-is-now-a-100m-business/","date":"2026-06-29","support":"TechCrunch says human-labelling firms Mercor, Surge and Scale AI compete with Arena \"for the same dollar\" in evaluation (paragraph 7); Labelbox's evaluation work has several qualified substitutes. Full date."},{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02-26","support":"Innodata's FY2025 10-K (Item 1, Competition) names Appen, CloudFactory, Surge AI, Invisible, Turing, Mercor, TELUS Digital and Scale AI as AI-data competitors; Labelbox work has many qualified homes. Date: EDGAR filing date."},{"url":"https://theaiinsider.tech/2026/10/15/labelbox-closes-100m-in-series-e-funding-to-advance-reinforcement-learning-and-enterprise-ai-agents/","date":"2026-10-15","support":"Labelbox says Horizon provides RL environments, expert feedback and evaluation infrastructure (paragraph 5), one of several such offerings. Full date."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02-26","support":"Innodata's 10-K (Item 1A) says customer agreements are generally terminable on about 30-90 days' notice, including on selecting another provider; Labelbox volume can be moved on that timescale. Date: EDGAR filing date."},{"url":"https://fortune.com/2025/06/19/openai-is-phasing-out-scale-ai-work-following-startups-meta-deal/","date":"2025-06-19","support":"OpenAI phased out Scale AI work within days of the Meta deal and relied increasingly on other providers such as Mercor (paragraphs 1, 9); Labelbox's expert-data work is similarly portable. Full date."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://techcrunch.com/2026/06/29/arena-the-ai-leaderboard-everyone-uses-is-now-a-100m-business/","date":"2026-06-29","support":"TechCrunch says human-labelling firms Mercor, Surge and Scale AI compete with Arena \"for the same dollar\" in evaluation (paragraph 7); Labelbox's evaluation work has several qualified substitutes. Full date."}]}},"data-tooling":{"alternatives":{"level":1,"evidence":[{"url":"https://lightly.ai/blog/the-10-best-encord-alternatives-in-2026-a-practical-guide-for-ml-team","date":"2026-05-18","support":"A May 2026 vendor guide lists ten labelling platforms (Labelbox, SuperAnnotate, Scale AI, V7, CVAT, Label Studio, Roboflow, Kili, Voxel51, Lightly) plus Dataloop as alternatives; many substitutes for Labelbox. Full date; author is a rival vendor."},{"url":"https://github.com/HumanSignal/label-studio","date":"undated","support":"Label Studio is an open-source, multi-type data labelling tool (text, audio, images, video, time series) with about 28.5k GitHub stars; a free substitute for Labelbox's software. Page undated."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://aws.amazon.com/about-aws/whats-new/2026/06/aws-service-availability/","date":"2026-06","support":"AWS ended SageMaker Ground Truth Plus on 30 June 2026 and closed Ground Truth to new customers from 30 July 2026: labelling-software users are expected to migrate within weeks. Relevant to Labelbox. Month taken from the URL path."},{"url":"https://docs.supervisely.com/import-and-export/migrations/migration-labelbox","date":"undated","support":"Supervisely offers a tool that copies multiple Labelbox projects into its platform; ready migration tooling keeps a switch away from Labelbox-type platforms to weeks. Page undated."},{"url":"https://lightly.ai/blog/the-10-best-encord-alternatives-in-2026-a-practical-guide-for-ml-team","date":"2026-05-18","support":"The guide advises \"Run the top two on a representative slice before you commit\", a short trial-based switch for platforms like Labelbox. Full date; author is a rival vendor."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://github.com/HumanSignal/label-studio","date":"undated","support":"Label Studio is an open-source, multi-type data labelling tool (text, audio, images, video, time series) with about 28.5k GitHub stars; a free substitute for Labelbox's software. Page undated."},{"url":"https://theaiinsider.tech/2026/10/15/labelbox-closes-100m-in-series-e-funding-to-advance-reinforcement-learning-and-enterprise-ai-agents/","date":"2026-10-15","support":"Labelbox says it has 400+ customers including over 90% of leading US AI labs (paragraph 2), but rival platforms already serve the same work. Full date."}]}}}},"b":{"schema":"5lc-index-rating/1","key":"models-data-evaluation","entityId":"labelbox","rater":"b","model":"claude-sonnet-5-5","ratedAt":"2026-10-22","functions":{"expert-data":{"alternatives":{"level":1,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02","support":"Competition section names nine AI-data specialists (including Appen, CloudFactory, Surge, Invisible, Turing, Mercor, TELUS Digital and Scale) plus large services firms; Labelbox is one of many. Date: month only, taken from the 20 Feb 2026 cover-page share count for the FY2025 filing."},{"url":"https://time.com/collection/time100-ai/2026/garrett-lord/","date":"2026-08-27","support":"Handshake AI says it works with eight frontier labs (27 Aug 2026), so several expert-data vendors are already qualified at labs."},{"url":"https://nebius.com/newsroom/nebius-welcomes-bezos-expeditions-as-lead-investor-in-ai-data-business-toloka","date":"2025-05-07","support":"Toloka lists Amazon, Anthropic, Microsoft, poolside, Recraft and Shopify as customers (7 May 2025), another vendor qualified at major buyers."},{"url":"https://www.labelbox.com/","date":"undated","support":"Labelbox says its Alignerr network has 2.6M+ skill-tested experts across 200+ domains, one of several large expert pools (page undated; the footer shows only a 2026 copyright)."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://www.storyboard18.com/how-it-works/google-to-sever-ties-with-scale-ai-following-metas-significant-stake-70359.htm","date":"2025-06-16","support":"Google reportedly planned to end about $200M a year of Scale spend after Meta's stake; Handshake's CEO said demand 'tripled overnight' and Labelbox expected hundreds of millions in new revenue (report of 16 Jun 2025)."},{"url":"https://siliconangle.com/2025/07/01/data-labeling-startup-surge-ai-seeking-1b-first-capital-raise-reports-say/","date":"2025-07-01","support":"Surge reportedly won former Scale customers including OpenAI and Google by 1 July 2025, under three weeks after Meta's stake was confirmed on 13 June (SiliconANGLE, citing reports)."},{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02","support":"Innodata says its customer agreements are generally terminable on 30 to 90 days' notice and do not commit customers to future purchases (risk factors). Date: month only."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://techcrunch.com/2026/07/08/these-ai-startups-are-growing-revenue-at-faster-and-faster-rates/","date":"2026-07-08","support":"Mercor says (CEO, TechCrunch 8 Jul 2026) gross annualised revenue rose from $500M (Sept 2025) to $1bn and then over $2bn (June 2026); Handshake and micro1 report about $1bn each (company statements), so several vendors already operate at the billion-dollar scale."},{"url":"https://newsletter.semianalysis.com/p/rl-environments-and-rl-for-science","date":"2026-01-06","support":"Multiple labs 'mostly stopped contracting' Scale after Meta's stake and the work was absorbed by others such as Surge; the article names no capability shortfall in the replacements (6 Jan 2026)."}]}},"evaluation":{"alternatives":{"level":1,"evidence":[{"url":"https://www.labelbox.com/","date":"undated","support":"Labelbox says over 90% of leading US labs 'post-train and evaluate' on it and cites Meta and Google Cloud evaluation studies (company claim) (page undated; the footer shows only a 2026 copyright)."},{"url":"https://labs.scale.com/leaderboard","date":"undated","support":"Scale Labs publishes 19 active leaderboards (page undated; the footer shows only a 2026 copyright)."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://mercor.com/blog/introducing-apex-ai-productivity-index/","date":"2025-10-01","support":"Mercor built its APEX v1 test set with about 100 experts in four professions (200 of 300 prompts approved), suggesting a comparable expert-written set is a matter of months (my inference, 1 Oct 2025)."},{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02","support":"Innodata says its customer agreements are generally terminable on 30 to 90 days' notice and do not commit customers to future purchases (risk factors). Date: month only."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://scale.com/blog/scales-next-era-building-for-2026","date":"2026-01-22","support":"Scale's lab says its lab introduced 15 new benchmarks and published over 450 evaluations of more than 50 models, a scale other vendors also reach (company blog, 22 Jan 2026)."},{"url":"https://welodata.ai/","date":"undated","support":"Welo Data lists evaluations, benchmarking, model selection and red teaming among its services (page undated; the footer shows only a 2026 copyright)."}]}},"data-tooling":{"alternatives":{"level":1,"evidence":[{"url":"https://github.com/HumanSignal/label-studio","date":"undated","support":"Label Studio is an Apache-2.0 open-source labelling tool with about 28.5k GitHub stars covering images, audio, text, video and time series (page undated; the footer shows only a 2026 copyright)."},{"url":"https://github.com/cvat-ai/cvat","date":"undated","support":"CVAT is an MIT-licensed annotation platform for image, video, audio and 3D data with about 17k GitHub stars (page undated; the footer shows only a 2026 copyright)."},{"url":"https://pulse2.com/encord-60-million-series-c-raised-to-scale-ai-native-data-infrastructure","date":"2026-02-27","support":"Encord sells a commercial multimodal annotation and curation platform to 300+ AI teams (press release, 27 Feb 2026)."},{"url":"https://www.vcaonline.com/news/2025071524/superannotate-raises-13-5-million-in-new-funding-from-dell-technologies-capital/","date":"2025-07-15","support":"SuperAnnotate sells a commercial annotation and evaluation platform to brands such as Databricks, Canva and Motorola Solutions (15 Jul 2025, company claim)."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://www.vcaonline.com/news/2025071524/superannotate-raises-13-5-million-in-new-funding-from-dell-technologies-capital/","date":"2025-07-15","support":"SuperAnnotate says its software revenue grew 5x in 2024 with customers such as Databricks and Canva, i.e. buyers adopt a platform quickly (15 Jul 2025, company claim; evidence thin)."},{"url":"https://github.com/HumanSignal/label-studio","date":"undated","support":"Open-source tools can be self-hosted and imported to without licence or procurement, shortening any switch (page undated; the footer shows only a 2026 copyright)."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://github.com/cvat-ai/cvat","date":"undated","support":"CVAT's MIT-licensed core already covers bounding boxes, polygons, masks, keypoints, cuboids and 3D point clouds (page undated; the footer shows only a 2026 copyright)."},{"url":"https://github.com/HumanSignal/label-studio","date":"undated","support":"Label Studio covers images, audio, text, video and time series with Apache-2.0 licensing (page undated; the footer shows only a 2026 copyright)."}]}}}}},"transperfect-dataforce":{"a":{"schema":"5lc-index-rating/1","key":"models-data-evaluation","entityId":"transperfect-dataforce","rater":"a","model":"claude-opus-5-5","ratedAt":"2026-10-22","functions":{"labelling-multimodal":{"alternatives":{"level":1,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02-26","support":"Innodata's FY2025 10-K (Item 1, Competition) names Appen, CloudFactory, Surge AI, Invisible, Turing, Mercor, TELUS Digital and Scale AI as AI-data competitors; TransPerfect DataForce work has many qualified homes. Date: EDGAR filing date."},{"url":"https://www.sec.gov/Archives/edgar/data/1829864/000182986426000012/task-20251231.htm","date":"2026-03-05","support":"TaskUs's FY2025 10-K (Item 1, Our Competition) calls AI data services a \"large, rapidly changing, and fragmented global market\"; many vendors can take TransPerfect DataForce's labelling work. Date: EDGAR filing date."}]},"switchingTime":{"level":2,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/1829864/000182986426000012/task-20251231.htm","date":"2026-03-05","support":"TaskUs's 10-K (Item 1A, labour risks) says hiring and training staff for increased demand \"takes time\"; absorbing TransPerfect DataForce's large programmes would take months, not weeks. Date: EDGAR filing date."},{"url":"https://www.searchenginejournal.com/google-terminates-contract-with-appen-for-search-quality-raters/506006/","date":"2024-01-21","support":"Google gave Appen notice on 20 January 2024 and ended its services by 19 March 2024 (paragraphs 1-2), a two-month exit; TransPerfect DataForce's volume could move similarly. Full date."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/1829864/000182986426000012/task-20251231.htm","date":"2026-03-05","support":"TaskUs's 10-K (Item 1) puts AI Services at 18% of US$1,183.5m 2025 revenue (about US$213m, 5LayerCake arithmetic), so rivals of TransPerfect DataForce already run labelling at large scale. Date: EDGAR filing date."}]}}}},"b":{"schema":"5lc-index-rating/1","key":"models-data-evaluation","entityId":"transperfect-dataforce","rater":"b","model":"claude-sonnet-5-5","ratedAt":"2026-10-22","functions":{"labelling-multimodal":{"alternatives":{"level":1,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02","support":"Competition section names nine AI-data specialists (including Appen, CloudFactory, Surge, Invisible, Turing, Mercor, TELUS Digital and Scale) plus large services firms; DataForce is one of many. Date: month only, taken from the 20 Feb 2026 cover-page share count for the FY2025 filing."},{"url":"https://www.dataforce.ai/","date":"undated","support":"DataForce lists over one million contributors and 200+ languages (page undated; the footer shows only a 2026 copyright)."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://announcements.asx.com.au/asxpdf/20250226/pdf/06fyjdglv66466.pdf","date":"2025-02-26","support":"Google's termination of its Appen contract in Q1 FY24 cut Appen's FY24 group revenue 14.2% to $234.3M, showing a large buyer can withdraw a labelling programme within a quarter (ASX, 26 Feb 2025)."},{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02","support":"Innodata says its customer agreements are generally terminable on 30 to 90 days' notice and do not commit customers to future purchases (risk factors). Date: month only."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://www.investegate.co.uk/announcement/rns/rws-holdings--rws/half-year-financial-report/9612324","date":"2026-06-11","support":"RWS TrainAI revenue (Generate segment) grew 44.7% to GBP 99.4M in the half to March 2026, with a new global technology client won, so incumbents are adding programmes and volume has somewhere to go (11 Jun 2026)."},{"url":"https://www.sec.gov/Archives/edgar/data/0000868675/000110465926088975/tm2615221d2_ex99-1.htm","date":"2026-07-31","support":"TELUS Digital's AI and data solutions line is shrinking on client ramp-downs, which suggests spare capacity among incumbents (31 Jul 2026)."}]}}}}},"toloka":{"a":{"schema":"5lc-index-rating/1","key":"models-data-evaluation","entityId":"toloka","rater":"a","model":"claude-opus-5-5","ratedAt":"2026-10-22","functions":{"expert-data":{"alternatives":{"level":1,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02-26","support":"Innodata's FY2025 10-K (Item 1, Competition) names Appen, CloudFactory, Surge AI, Invisible, Turing, Mercor, TELUS Digital and Scale AI as AI-data competitors; Toloka work has many qualified homes. Date: EDGAR filing date."},{"url":"https://nebius.com/newsroom/nebius-welcomes-bezos-expeditions-as-lead-investor-in-ai-data-business-toloka","date":"2025-05-07","support":"Nebius says Toloka serves Amazon, Anthropic, Microsoft, poolside, Recraft and Shopify (paragraph 12), customers that also use other vendors. Full date."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://fortune.com/2025/06/19/openai-is-phasing-out-scale-ai-work-following-startups-meta-deal/","date":"2025-06-19","support":"OpenAI phased out Scale AI work within days of the Meta deal and relied increasingly on other providers such as Mercor (paragraphs 1, 9); Toloka's expert-data work is similarly portable. Full date."},{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02-26","support":"Innodata's 10-K (Item 1A) says customer agreements are generally terminable on about 30-90 days' notice, including on selecting another provider; Toloka volume can be moved on that timescale. Date: EDGAR filing date."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://techcrunch.com/2026/09/22/snorkel-ai-triples-valuation-to-3-5b-as-demand-for-ai-training-data-booms/","date":"2026-09-22","support":"TechCrunch names Mercor (US$2bn gross), Handshake (US$1bn) and micro1 (US$500m) as scaled peers, and Snorkel at US$375m run-rate (company says) (paragraphs 4-5); several large alternatives to Toloka. Full date."}]}},"labelling-multimodal":{"alternatives":{"level":1,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02-26","support":"Innodata's FY2025 10-K (Item 1, Competition) names Appen, CloudFactory, Surge AI, Invisible, Turing, Mercor, TELUS Digital and Scale AI as AI-data competitors; Toloka work has many qualified homes. Date: EDGAR filing date."},{"url":"https://www.sec.gov/Archives/edgar/data/1829864/000182986426000012/task-20251231.htm","date":"2026-03-05","support":"TaskUs's FY2025 10-K (Item 1, Our Competition) calls AI data services a \"large, rapidly changing, and fragmented global market\"; many vendors can take Toloka's labelling work. Date: EDGAR filing date."},{"url":"https://nebius.com/newsroom/nebius-welcomes-bezos-expeditions-as-lead-investor-in-ai-data-business-toloka","date":"2025-05-07","support":"Toloka began as large-scale crowd-sourced labelling (paragraph 12), a widely supplied service. Full date."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://www.searchenginejournal.com/google-terminates-contract-with-appen-for-search-quality-raters/506006/","date":"2024-01-21","support":"Google gave Appen notice on 20 January 2024 and ended its services by 19 March 2024 (paragraphs 1-2), a two-month exit; Toloka's volume could move similarly. Full date."},{"url":"https://www.sec.gov/Archives/edgar/data/1829864/000182986426000012/task-20251231.htm","date":"2026-03-05","support":"TaskUs's 10-K (Item 1A) says clients can terminate for convenience or move work in-house or to competitors; Toloka's programmes are portable within weeks to months. Date: EDGAR filing date."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/1829864/000182986426000012/task-20251231.htm","date":"2026-03-05","support":"TaskUs's 10-K (Item 1) puts AI Services at 18% of US$1,183.5m 2025 revenue (about US$213m, 5LayerCake arithmetic), so rivals of Toloka already run labelling at large scale. Date: EDGAR filing date."}]}},"evaluation":{"alternatives":{"level":1,"evidence":[{"url":"https://techcrunch.com/2026/06/29/arena-the-ai-leaderboard-everyone-uses-is-now-a-100m-business/","date":"2026-06-29","support":"TechCrunch says human-labelling firms Mercor, Surge and Scale AI compete with Arena \"for the same dollar\" in evaluation (paragraph 7); Toloka's evaluation work has several qualified substitutes. Full date."},{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02-26","support":"Innodata's FY2025 10-K (Item 1, Competition) names Appen, CloudFactory, Surge AI, Invisible, Turing, Mercor, TELUS Digital and Scale AI as AI-data competitors; Toloka work has many qualified homes. Date: EDGAR filing date."},{"url":"https://nebius.com/newsroom/nebius-welcomes-bezos-expeditions-as-lead-investor-in-ai-data-business-toloka","date":"2025-05-07","support":"Nebius cites Toloka's agent safety evaluation and red-teaming expertise (paragraph 12), offered by several other vendors too. Full date."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02-26","support":"Innodata's 10-K (Item 1A) says customer agreements are generally terminable on about 30-90 days' notice, including on selecting another provider; Toloka volume can be moved on that timescale. Date: EDGAR filing date."},{"url":"https://fortune.com/2025/06/19/openai-is-phasing-out-scale-ai-work-following-startups-meta-deal/","date":"2025-06-19","support":"OpenAI phased out Scale AI work within days of the Meta deal and relied increasingly on other providers such as Mercor (paragraphs 1, 9); Toloka's expert-data work is similarly portable. Full date."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://techcrunch.com/2026/06/29/arena-the-ai-leaderboard-everyone-uses-is-now-a-100m-business/","date":"2026-06-29","support":"TechCrunch says human-labelling firms Mercor, Surge and Scale AI compete with Arena \"for the same dollar\" in evaluation (paragraph 7); Toloka's evaluation work has several qualified substitutes. Full date."}]}}}},"b":{"schema":"5lc-index-rating/1","key":"models-data-evaluation","entityId":"toloka","rater":"b","model":"claude-sonnet-5-5","ratedAt":"2026-10-22","functions":{"expert-data":{"alternatives":{"level":1,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02","support":"Competition section names nine AI-data specialists (including Appen, CloudFactory, Surge, Invisible, Turing, Mercor, TELUS Digital and Scale) plus large services firms; Toloka is one of many. Date: month only, taken from the 20 Feb 2026 cover-page share count for the FY2025 filing."},{"url":"https://time.com/collection/time100-ai/2026/garrett-lord/","date":"2026-08-27","support":"Handshake AI says it works with eight frontier labs (27 Aug 2026), so several expert-data vendors are already qualified at labs."},{"url":"https://nebius.com/newsroom/nebius-welcomes-bezos-expeditions-as-lead-investor-in-ai-data-business-toloka","date":"2025-05-07","support":"Toloka lists Amazon, Anthropic, Microsoft, poolside, Recraft and Shopify as customers (7 May 2025), another vendor qualified at major buyers."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://www.storyboard18.com/how-it-works/google-to-sever-ties-with-scale-ai-following-metas-significant-stake-70359.htm","date":"2025-06-16","support":"Google reportedly planned to end about $200M a year of Scale spend after Meta's stake; Handshake's CEO said demand 'tripled overnight' and Labelbox expected hundreds of millions in new revenue (report of 16 Jun 2025)."},{"url":"https://siliconangle.com/2025/07/01/data-labeling-startup-surge-ai-seeking-1b-first-capital-raise-reports-say/","date":"2025-07-01","support":"Surge reportedly won former Scale customers including OpenAI and Google by 1 July 2025, under three weeks after Meta's stake was confirmed on 13 June (SiliconANGLE, citing reports)."},{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02","support":"Innodata says its customer agreements are generally terminable on 30 to 90 days' notice and do not commit customers to future purchases (risk factors). Date: month only."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://techcrunch.com/2026/07/08/these-ai-startups-are-growing-revenue-at-faster-and-faster-rates/","date":"2026-07-08","support":"Mercor says (CEO, TechCrunch 8 Jul 2026) gross annualised revenue rose from $500M (Sept 2025) to $1bn and then over $2bn (June 2026); Handshake and micro1 report about $1bn each (company statements), so several vendors already operate at the billion-dollar scale."},{"url":"https://newsletter.semianalysis.com/p/rl-environments-and-rl-for-science","date":"2026-01-06","support":"Multiple labs 'mostly stopped contracting' Scale after Meta's stake and the work was absorbed by others such as Surge; the article names no capability shortfall in the replacements (6 Jan 2026)."}]}},"labelling-multimodal":{"alternatives":{"level":1,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02","support":"Competition section names nine AI-data specialists (including Appen, CloudFactory, Surge, Invisible, Turing, Mercor, TELUS Digital and Scale) plus large services firms; Toloka is one of many. Date: month only, taken from the 20 Feb 2026 cover-page share count for the FY2025 filing."},{"url":"https://www.sec.gov/Archives/edgar/data/0001829864/000182986426000147/earningsreleaseex991q22026.htm","date":"2026-08-05","support":"TaskUs reports AI Services as its fastest-growing line, up 25.8% year over year in Q2 2026, a large listed provider of AI data work (release of 5 Aug 2026)."},{"url":"https://announcements.asx.com.au/asxpdf/20260827/pdf/0739d4q7lyh3w2.pdf","date":"2026-08-27","support":"Appen China's annualised run-rate passed $175M in June 2026 and Appen Global is rebuilding; Appen is a further large vendor (ASX, 27 Aug 2026)."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://announcements.asx.com.au/asxpdf/20250226/pdf/06fyjdglv66466.pdf","date":"2025-02-26","support":"Google's termination of its Appen contract in Q1 FY24 cut Appen's FY24 group revenue 14.2% to $234.3M, showing a large buyer can withdraw a labelling programme within a quarter (ASX, 26 Feb 2025)."},{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02","support":"Innodata says its customer agreements are generally terminable on 30 to 90 days' notice and do not commit customers to future purchases (risk factors). Date: month only."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://www.investegate.co.uk/announcement/rns/rws-holdings--rws/half-year-financial-report/9612324","date":"2026-06-11","support":"RWS TrainAI revenue (Generate segment) grew 44.7% to GBP 99.4M in the half to March 2026, with a new global technology client won, so incumbents are adding programmes and volume has somewhere to go (11 Jun 2026)."},{"url":"https://www.sec.gov/Archives/edgar/data/0000868675/000110465926088975/tm2615221d2_ex99-1.htm","date":"2026-07-31","support":"TELUS Digital's AI and data solutions line is shrinking on client ramp-downs, which suggests spare capacity among incumbents (31 Jul 2026)."}]}},"evaluation":{"alternatives":{"level":1,"evidence":[{"url":"https://nebius.com/newsroom/nebius-welcomes-bezos-expeditions-as-lead-investor-in-ai-data-business-toloka","date":"2025-05-07","support":"Toloka itself offers agent safety evaluation and red teaming to Amazon, Anthropic, Microsoft and others (7 May 2025); Prolific and Welo sell similar services."},{"url":"https://www.prolific.com/","date":"undated","support":"Prolific sells human feedback for preference tuning, safety evals and benchmarks and lists a HUMAINE leaderboard (page undated; the footer shows only a 2026 copyright)."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://mercor.com/blog/introducing-apex-ai-productivity-index/","date":"2025-10-01","support":"Mercor built its APEX v1 test set with about 100 experts in four professions (200 of 300 prompts approved), suggesting a comparable expert-written set is a matter of months (my inference, 1 Oct 2025)."},{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02","support":"Innodata says its customer agreements are generally terminable on 30 to 90 days' notice and do not commit customers to future purchases (risk factors). Date: month only."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://scale.com/blog/scales-next-era-building-for-2026","date":"2026-01-22","support":"Scale's lab says its lab introduced 15 new benchmarks and published over 450 evaluations of more than 50 models, a scale other vendors also reach (company blog, 22 Jan 2026)."},{"url":"https://welodata.ai/","date":"undated","support":"Welo Data lists evaluations, benchmarking, model selection and red teaming among its services (page undated; the footer shows only a 2026 copyright)."}]}}}}},"fleet-ai":{"a":{"schema":"5lc-index-rating/1","key":"models-data-evaluation","entityId":"fleet-ai","rater":"a","model":"claude-opus-5-5","ratedAt":"2026-10-22","functions":{"rl-environments":{"alternatives":{"level":1,"evidence":[{"url":"https://newsletter.semianalysis.com/p/rl-environments-and-rl-for-science","date":"2026-01-06","support":"SemiAnalysis counts \"more than 35 companies\" building RL environments, mostly seed-stage, each serving 1-3 customers (vendor section); Fleet faces many substitutes. Full date."},{"url":"https://newsletter.semianalysis.com/p/rl-environments-and-rl-for-science","date":"2026-01-06","support":"SemiAnalysis names Fleet among environment builders such as Habitat, DeepTune, Vmax, Turing, Mechanize and Bespoke Labs (vendor section). Full date."}]},"switchingTime":{"level":2,"evidence":[{"url":"https://newsletter.semianalysis.com/p/rl-environments-and-rl-for-science","date":"2026-01-06","support":"SemiAnalysis says most environment vendors sell under exclusive contracts and UI-gym sites cost about US$20,000 each as one-time purchases (vendor and pricing sections); replacing Fleet's pipeline means commissioning new builds. Full date."},{"url":"https://techcrunch.com/2025/09/16/silicon-valley-bets-big-on-environments-to-train-ai-agents/","date":"2025-09-16","support":"Ross Taylor warns \"people are underestimating how difficult it is to scale environments\" (paragraph 32); a lab replacing Fleet's environment output would need months to re-commission it. Full date."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://epoch.ai/gradient-updates/state-of-rl-envs","date":"2026-01-12","support":"Epoch's RL-environment FAQ lists specialised startups, data providers (Mercor, Surge, Handshake, Turing), product companies and labs' in-house teams as sources ('Which companies build RL Environments?'); many alternatives to Fleet. Full date."}]}}}},"b":{"schema":"5lc-index-rating/1","key":"models-data-evaluation","entityId":"fleet-ai","rater":"b","model":"claude-sonnet-5-5","ratedAt":"2026-10-22","functions":{"rl-environments":{"alternatives":{"level":1,"evidence":[{"url":"https://sacra.com/c/fleet","date":"undated","support":"Sacra lists Surge, Mercor and Turing, lab in-housing and open-source standards (BrowserGym, OpenEnv) as Fleet's competition (page undated; the footer shows only a 2026 copyright)."},{"url":"https://newsletter.semianalysis.com/p/rl-environments-and-rl-for-science","date":"2026-01-06","support":"More than 35 companies sell RL environments (mostly seed-stage) and Anthropic likely works with more than a dozen of them (SemiAnalysis, 6 Jan 2026)."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://sacra.com/c/fleet","date":"undated","support":"Sacra notes Fleet's revenue depends on a few labs that could bring environment work in-house and reflects one quarter of lab demand (page undated; the footer shows only a 2026 copyright)."},{"url":"https://epoch.ai/gradient-updates/state-of-rl-envs","date":"2026-01-12","support":"Environment contracts are often six to seven figures per quarter and can be non-exclusive, so labs can add or swap vendors quarter by quarter; Epoch also reports 'substantially more' in-housing (12 Jan 2026)."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://sacra.com/c/fleet","date":"undated","support":"Open-source standards are 'commoditizing interfaces', and data vendors can upsell environments to existing lab relationships (page undated; the footer shows only a 2026 copyright)."},{"url":"https://epoch.ai/gradient-updates/state-of-rl-envs","date":"2026-01-12","support":"Quality at scale is the main bottleneck for all vendors; interviewees rank robustness to reward hacking first and say domain knowledge matters more than ML skill (12 Jan 2026). No vendor is described as uniquely capable."}]}}}}},"uber-ai-solutions":{"a":{"schema":"5lc-index-rating/1","key":"models-data-evaluation","entityId":"uber-ai-solutions","rater":"a","model":"claude-opus-5-5","ratedAt":"2026-10-22","functions":{"labelling-multimodal":{"alternatives":{"level":1,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02-26","support":"Innodata's FY2025 10-K (Item 1, Competition) names Appen, CloudFactory, Surge AI, Invisible, Turing, Mercor, TELUS Digital and Scale AI as AI-data competitors; Uber AI Solutions work has many qualified homes. Date: EDGAR filing date."},{"url":"https://www.sec.gov/Archives/edgar/data/1829864/000182986426000012/task-20251231.htm","date":"2026-03-05","support":"TaskUs's FY2025 10-K (Item 1, Our Competition) calls AI data services a \"large, rapidly changing, and fragmented global market\"; many vendors can take Uber AI Solutions's labelling work. Date: EDGAR filing date."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://www.searchenginejournal.com/google-terminates-contract-with-appen-for-search-quality-raters/506006/","date":"2024-01-21","support":"Google gave Appen notice on 20 January 2024 and ended its services by 19 March 2024 (paragraphs 1-2), a two-month exit; Uber AI Solutions's volume could move similarly. Full date."},{"url":"https://www.sec.gov/Archives/edgar/data/1829864/000182986426000012/task-20251231.htm","date":"2026-03-05","support":"TaskUs's 10-K (Item 1A) says clients can terminate for convenience or move work in-house or to competitors; Uber AI Solutions's programmes are portable within weeks to months. Date: EDGAR filing date."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/1829864/000182986426000012/task-20251231.htm","date":"2026-03-05","support":"TaskUs's 10-K (Item 1) puts AI Services at 18% of US$1,183.5m 2025 revenue (about US$213m, 5LayerCake arithmetic), so rivals of Uber AI Solutions already run labelling at large scale. Date: EDGAR filing date."}]}}}},"b":{"schema":"5lc-index-rating/1","key":"models-data-evaluation","entityId":"uber-ai-solutions","rater":"b","model":"claude-sonnet-5-5","ratedAt":"2026-10-22","functions":{"labelling-multimodal":{"alternatives":{"level":1,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02","support":"Competition section names nine AI-data specialists (including Appen, CloudFactory, Surge, Invisible, Turing, Mercor, TELUS Digital and Scale) plus large services firms; Uber AI Solutions is one of many. Date: month only, taken from the 20 Feb 2026 cover-page share count for the FY2025 filing."},{"url":"https://startupfortune.com/uber-fires-the-two-executives-who-built-its-ai-data-labeling-bet/","date":"2026-07-01","support":"Mercor, Turing, Invisible and Surge are fighting for the same work; Uber has disclosed no revenue and two senior leaders left in June 2026 (1 Jul 2026)."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://announcements.asx.com.au/asxpdf/20250226/pdf/06fyjdglv66466.pdf","date":"2025-02-26","support":"Google's termination of its Appen contract in Q1 FY24 cut Appen's FY24 group revenue 14.2% to $234.3M, showing a large buyer can withdraw a labelling programme within a quarter (ASX, 26 Feb 2025)."},{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02","support":"Innodata says its customer agreements are generally terminable on 30 to 90 days' notice and do not commit customers to future purchases (risk factors). Date: month only."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://www.investegate.co.uk/announcement/rns/rws-holdings--rws/half-year-financial-report/9612324","date":"2026-06-11","support":"RWS TrainAI revenue (Generate segment) grew 44.7% to GBP 99.4M in the half to March 2026, with a new global technology client won, so incumbents are adding programmes and volume has somewhere to go (11 Jun 2026)."},{"url":"https://www.sec.gov/Archives/edgar/data/0000868675/000110465926088975/tm2615221d2_ex99-1.htm","date":"2026-07-31","support":"TELUS Digital's AI and data solutions line is shrinking on client ramp-downs, which suggests spare capacity among incumbents (31 Jul 2026)."}]}}}}},"welo-data":{"a":{"schema":"5lc-index-rating/1","key":"models-data-evaluation","entityId":"welo-data","rater":"a","model":"claude-opus-5-5","ratedAt":"2026-10-22","functions":{"labelling-multimodal":{"alternatives":{"level":1,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02-26","support":"Innodata's FY2025 10-K (Item 1, Competition) names Appen, CloudFactory, Surge AI, Invisible, Turing, Mercor, TELUS Digital and Scale AI as AI-data competitors; Welo Data work has many qualified homes. Date: EDGAR filing date."},{"url":"https://www.sec.gov/Archives/edgar/data/1829864/000182986426000012/task-20251231.htm","date":"2026-03-05","support":"TaskUs's FY2025 10-K (Item 1, Our Competition) calls AI data services a \"large, rapidly changing, and fragmented global market\"; many vendors can take Welo Data's labelling work. Date: EDGAR filing date."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://www.searchenginejournal.com/google-terminates-contract-with-appen-for-search-quality-raters/506006/","date":"2024-01-21","support":"Google gave Appen notice on 20 January 2024 and ended its services by 19 March 2024 (paragraphs 1-2), a two-month exit; Welo Data's volume could move similarly. Full date."},{"url":"https://www.sec.gov/Archives/edgar/data/1829864/000182986426000012/task-20251231.htm","date":"2026-03-05","support":"TaskUs's 10-K (Item 1A) says clients can terminate for convenience or move work in-house or to competitors; Welo Data's programmes are portable within weeks to months. Date: EDGAR filing date."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/1829864/000182986426000012/task-20251231.htm","date":"2026-03-05","support":"TaskUs's 10-K (Item 1) puts AI Services at 18% of US$1,183.5m 2025 revenue (about US$213m, 5LayerCake arithmetic), so rivals of Welo Data already run labelling at large scale. Date: EDGAR filing date."}]}}}},"b":{"schema":"5lc-index-rating/1","key":"models-data-evaluation","entityId":"welo-data","rater":"b","model":"claude-sonnet-5-5","ratedAt":"2026-10-22","functions":{"labelling-multimodal":{"alternatives":{"level":1,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02","support":"Competition section names nine AI-data specialists (including Appen, CloudFactory, Surge, Invisible, Turing, Mercor, TELUS Digital and Scale) plus large services firms; Welo Data is one of many. Date: month only, taken from the 20 Feb 2026 cover-page share count for the FY2025 filing."},{"url":"https://welodata.ai/","date":"undated","support":"Welo Data lists 500k+ curated experts and 155+ locales (page undated; the footer shows only a 2026 copyright)."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://announcements.asx.com.au/asxpdf/20250226/pdf/06fyjdglv66466.pdf","date":"2025-02-26","support":"Google's termination of its Appen contract in Q1 FY24 cut Appen's FY24 group revenue 14.2% to $234.3M, showing a large buyer can withdraw a labelling programme within a quarter (ASX, 26 Feb 2025)."},{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02","support":"Innodata says its customer agreements are generally terminable on 30 to 90 days' notice and do not commit customers to future purchases (risk factors). Date: month only."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://www.investegate.co.uk/announcement/rns/rws-holdings--rws/half-year-financial-report/9612324","date":"2026-06-11","support":"RWS TrainAI revenue (Generate segment) grew 44.7% to GBP 99.4M in the half to March 2026, with a new global technology client won, so incumbents are adding programmes and volume has somewhere to go (11 Jun 2026)."},{"url":"https://www.sec.gov/Archives/edgar/data/0000868675/000110465926088975/tm2615221d2_ex99-1.htm","date":"2026-07-31","support":"TELUS Digital's AI and data solutions line is shrinking on client ramp-downs, which suggests spare capacity among incumbents (31 Jul 2026)."}]}}}}},"imerit":{"a":{"schema":"5lc-index-rating/1","key":"models-data-evaluation","entityId":"imerit","rater":"a","model":"claude-opus-5-5","ratedAt":"2026-10-22","functions":{"labelling-multimodal":{"alternatives":{"level":1,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02-26","support":"Innodata's FY2025 10-K (Item 1, Competition) names Appen, CloudFactory, Surge AI, Invisible, Turing, Mercor, TELUS Digital and Scale AI as AI-data competitors; iMerit work has many qualified homes. Date: EDGAR filing date."},{"url":"https://www.sec.gov/Archives/edgar/data/1829864/000182986426000012/task-20251231.htm","date":"2026-03-05","support":"TaskUs's FY2025 10-K (Item 1, Our Competition) calls AI data services a \"large, rapidly changing, and fragmented global market\"; many vendors can take iMerit's labelling work. Date: EDGAR filing date."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://www.searchenginejournal.com/google-terminates-contract-with-appen-for-search-quality-raters/506006/","date":"2024-01-21","support":"Google gave Appen notice on 20 January 2024 and ended its services by 19 March 2024 (paragraphs 1-2), a two-month exit; iMerit's volume could move similarly. Full date."},{"url":"https://www.sec.gov/Archives/edgar/data/1829864/000182986426000012/task-20251231.htm","date":"2026-03-05","support":"TaskUs's 10-K (Item 1A) says clients can terminate for convenience or move work in-house or to competitors; iMerit's programmes are portable within weeks to months. Date: EDGAR filing date."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/1829864/000182986426000012/task-20251231.htm","date":"2026-03-05","support":"TaskUs's 10-K (Item 1) puts AI Services at 18% of US$1,183.5m 2025 revenue (about US$213m, 5LayerCake arithmetic), so rivals of iMerit already run labelling at large scale. Date: EDGAR filing date."}]}},"data-tooling":{"alternatives":{"level":1,"evidence":[{"url":"https://lightly.ai/blog/the-10-best-encord-alternatives-in-2026-a-practical-guide-for-ml-team","date":"2026-05-18","support":"A May 2026 vendor guide lists ten labelling platforms (Labelbox, SuperAnnotate, Scale AI, V7, CVAT, Label Studio, Roboflow, Kili, Voxel51, Lightly) plus Dataloop as alternatives; many substitutes for iMerit. Full date; author is a rival vendor."},{"url":"https://github.com/HumanSignal/label-studio","date":"undated","support":"Label Studio is an open-source, multi-type data labelling tool (text, audio, images, video, time series) with about 28.5k GitHub stars; a free substitute for iMerit's software. Page undated."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://aws.amazon.com/about-aws/whats-new/2026/06/aws-service-availability/","date":"2026-06","support":"AWS ended SageMaker Ground Truth Plus on 30 June 2026 and closed Ground Truth to new customers from 30 July 2026: labelling-software users are expected to migrate within weeks. Relevant to iMerit. Month taken from the URL path."},{"url":"https://docs.supervisely.com/import-and-export/migrations/migration-labelbox","date":"undated","support":"Supervisely offers a tool that copies multiple Labelbox projects into its platform; ready migration tooling keeps a switch away from iMerit-type platforms to weeks. Page undated."},{"url":"https://lightly.ai/blog/the-10-best-encord-alternatives-in-2026-a-practical-guide-for-ml-team","date":"2026-05-18","support":"The guide advises \"Run the top two on a representative slice before you commit\", a short trial-based switch for platforms like iMerit. Full date; author is a rival vendor."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://github.com/HumanSignal/label-studio","date":"undated","support":"Label Studio is an open-source, multi-type data labelling tool (text, audio, images, video, time series) with about 28.5k GitHub stars; a free substitute for iMerit's software. Page undated."},{"url":"https://lightly.ai/blog/the-10-best-encord-alternatives-in-2026-a-practical-guide-for-ml-team","date":"2026-05-18","support":"A May 2026 vendor guide lists ten labelling platforms (Labelbox, SuperAnnotate, Scale AI, V7, CVAT, Label Studio, Roboflow, Kili, Voxel51, Lightly) plus Dataloop as alternatives; many substitutes for iMerit. Full date; author is a rival vendor."}]}}}},"b":{"schema":"5lc-index-rating/1","key":"models-data-evaluation","entityId":"imerit","rater":"b","model":"claude-sonnet-5-5","ratedAt":"2026-10-22","functions":{"labelling-multimodal":{"alternatives":{"level":1,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02","support":"Competition section names nine AI-data specialists (including Appen, CloudFactory, Surge, Invisible, Turing, Mercor, TELUS Digital and Scale) plus large services firms; iMerit is one of many. Date: month only, taken from the 20 Feb 2026 cover-page share count for the FY2025 filing."},{"url":"https://sec.gov/Archives/edgar/data/1297989/000129798926000021/exlq226pressrelease.htm","date":"2026-07-28","support":"EXL guides to $28.0-32.0M of anticipated 2026 revenue from iMerit after the expected 31 Jul 2026 close, a mid-sized vendor beside many others (28 Jul 2026)."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://announcements.asx.com.au/asxpdf/20250226/pdf/06fyjdglv66466.pdf","date":"2025-02-26","support":"Google's termination of its Appen contract in Q1 FY24 cut Appen's FY24 group revenue 14.2% to $234.3M, showing a large buyer can withdraw a labelling programme within a quarter (ASX, 26 Feb 2025)."},{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02","support":"Innodata says its customer agreements are generally terminable on 30 to 90 days' notice and do not commit customers to future purchases (risk factors). Date: month only."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://www.investegate.co.uk/announcement/rns/rws-holdings--rws/half-year-financial-report/9612324","date":"2026-06-11","support":"RWS TrainAI revenue (Generate segment) grew 44.7% to GBP 99.4M in the half to March 2026, with a new global technology client won, so incumbents are adding programmes and volume has somewhere to go (11 Jun 2026)."},{"url":"https://www.sec.gov/Archives/edgar/data/0000868675/000110465926088975/tm2615221d2_ex99-1.htm","date":"2026-07-31","support":"TELUS Digital's AI and data solutions line is shrinking on client ramp-downs, which suggests spare capacity among incumbents (31 Jul 2026)."}]}},"data-tooling":{"alternatives":{"level":1,"evidence":[{"url":"https://github.com/HumanSignal/label-studio","date":"undated","support":"Label Studio is an Apache-2.0 open-source labelling tool with about 28.5k GitHub stars covering images, audio, text, video and time series (page undated; the footer shows only a 2026 copyright)."},{"url":"https://github.com/cvat-ai/cvat","date":"undated","support":"CVAT is an MIT-licensed annotation platform for image, video, audio and 3D data with about 17k GitHub stars (page undated; the footer shows only a 2026 copyright)."},{"url":"https://pulse2.com/encord-60-million-series-c-raised-to-scale-ai-native-data-infrastructure","date":"2026-02-27","support":"Encord sells a commercial multimodal annotation and curation platform to 300+ AI teams (press release, 27 Feb 2026)."},{"url":"https://www.vcaonline.com/news/2025071524/superannotate-raises-13-5-million-in-new-funding-from-dell-technologies-capital/","date":"2025-07-15","support":"SuperAnnotate sells a commercial annotation and evaluation platform to brands such as Databricks, Canva and Motorola Solutions (15 Jul 2025, company claim)."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://www.vcaonline.com/news/2025071524/superannotate-raises-13-5-million-in-new-funding-from-dell-technologies-capital/","date":"2025-07-15","support":"SuperAnnotate says its software revenue grew 5x in 2024 with customers such as Databricks and Canva, i.e. buyers adopt a platform quickly (15 Jul 2025, company claim; evidence thin)."},{"url":"https://github.com/HumanSignal/label-studio","date":"undated","support":"Open-source tools can be self-hosted and imported to without licence or procurement, shortening any switch (page undated; the footer shows only a 2026 copyright)."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://github.com/cvat-ai/cvat","date":"undated","support":"CVAT's MIT-licensed core already covers bounding boxes, polygons, masks, keypoints, cuboids and 3D point clouds (page undated; the footer shows only a 2026 copyright)."},{"url":"https://github.com/HumanSignal/label-studio","date":"undated","support":"Label Studio covers images, audio, text, video and time series with Apache-2.0 licensing (page undated; the footer shows only a 2026 copyright)."}]}}}}},"lionbridge":{"a":{"schema":"5lc-index-rating/1","key":"models-data-evaluation","entityId":"lionbridge","rater":"a","model":"claude-opus-5-5","ratedAt":"2026-10-22","functions":{"labelling-multimodal":{"alternatives":{"level":1,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02-26","support":"Innodata's FY2025 10-K (Item 1, Competition) names Appen, CloudFactory, Surge AI, Invisible, Turing, Mercor, TELUS Digital and Scale AI as AI-data competitors; Lionbridge work has many qualified homes. Date: EDGAR filing date."},{"url":"https://www.sec.gov/Archives/edgar/data/1829864/000182986426000012/task-20251231.htm","date":"2026-03-05","support":"TaskUs's FY2025 10-K (Item 1, Our Competition) calls AI data services a \"large, rapidly changing, and fragmented global market\"; many vendors can take Lionbridge's labelling work. Date: EDGAR filing date."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://www.searchenginejournal.com/google-terminates-contract-with-appen-for-search-quality-raters/506006/","date":"2024-01-21","support":"Google gave Appen notice on 20 January 2024 and ended its services by 19 March 2024 (paragraphs 1-2), a two-month exit; Lionbridge's volume could move similarly. Full date."},{"url":"https://www.sec.gov/Archives/edgar/data/1829864/000182986426000012/task-20251231.htm","date":"2026-03-05","support":"TaskUs's 10-K (Item 1A) says clients can terminate for convenience or move work in-house or to competitors; Lionbridge's programmes are portable within weeks to months. Date: EDGAR filing date."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/1829864/000182986426000012/task-20251231.htm","date":"2026-03-05","support":"TaskUs's 10-K (Item 1) puts AI Services at 18% of US$1,183.5m 2025 revenue (about US$213m, 5LayerCake arithmetic), so rivals of Lionbridge already run labelling at large scale. Date: EDGAR filing date."}]}}}},"b":{"schema":"5lc-index-rating/1","key":"models-data-evaluation","entityId":"lionbridge","rater":"b","model":"claude-sonnet-5-5","ratedAt":"2026-10-22","functions":{"labelling-multimodal":{"alternatives":{"level":1,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02","support":"Competition section names nine AI-data specialists (including Appen, CloudFactory, Surge, Invisible, Turing, Mercor, TELUS Digital and Scale) plus large services firms; Lionbridge is one of many. Date: month only, taken from the 20 Feb 2026 cover-page share count for the FY2025 filing."},{"url":"https://www.lionbridge.com/","date":"undated","support":"Lionbridge lists a community of more than half a million experts in AI data services (page undated; the footer shows only a 2026 copyright)."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://announcements.asx.com.au/asxpdf/20250226/pdf/06fyjdglv66466.pdf","date":"2025-02-26","support":"Google's termination of its Appen contract in Q1 FY24 cut Appen's FY24 group revenue 14.2% to $234.3M, showing a large buyer can withdraw a labelling programme within a quarter (ASX, 26 Feb 2025)."},{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02","support":"Innodata says its customer agreements are generally terminable on 30 to 90 days' notice and do not commit customers to future purchases (risk factors). Date: month only."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://www.investegate.co.uk/announcement/rns/rws-holdings--rws/half-year-financial-report/9612324","date":"2026-06-11","support":"RWS TrainAI revenue (Generate segment) grew 44.7% to GBP 99.4M in the half to March 2026, with a new global technology client won, so incumbents are adding programmes and volume has somewhere to go (11 Jun 2026)."},{"url":"https://www.sec.gov/Archives/edgar/data/0000868675/000110465926088975/tm2615221d2_ex99-1.htm","date":"2026-07-31","support":"TELUS Digital's AI and data solutions line is shrinking on client ramp-downs, which suggests spare capacity among incumbents (31 Jul 2026)."}]}}}}},"lxt":{"a":{"schema":"5lc-index-rating/1","key":"models-data-evaluation","entityId":"lxt","rater":"a","model":"claude-opus-5-5","ratedAt":"2026-10-22","functions":{"labelling-multimodal":{"alternatives":{"level":1,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02-26","support":"Innodata's FY2025 10-K (Item 1, Competition) names Appen, CloudFactory, Surge AI, Invisible, Turing, Mercor, TELUS Digital and Scale AI as AI-data competitors; LXT work has many qualified homes. Date: EDGAR filing date."},{"url":"https://www.sec.gov/Archives/edgar/data/1829864/000182986426000012/task-20251231.htm","date":"2026-03-05","support":"TaskUs's FY2025 10-K (Item 1, Our Competition) calls AI data services a \"large, rapidly changing, and fragmented global market\"; many vendors can take LXT's labelling work. Date: EDGAR filing date."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://www.searchenginejournal.com/google-terminates-contract-with-appen-for-search-quality-raters/506006/","date":"2024-01-21","support":"Google gave Appen notice on 20 January 2024 and ended its services by 19 March 2024 (paragraphs 1-2), a two-month exit; LXT's volume could move similarly. Full date."},{"url":"https://www.sec.gov/Archives/edgar/data/1829864/000182986426000012/task-20251231.htm","date":"2026-03-05","support":"TaskUs's 10-K (Item 1A) says clients can terminate for convenience or move work in-house or to competitors; LXT's programmes are portable within weeks to months. Date: EDGAR filing date."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/1829864/000182986426000012/task-20251231.htm","date":"2026-03-05","support":"TaskUs's 10-K (Item 1) puts AI Services at 18% of US$1,183.5m 2025 revenue (about US$213m, 5LayerCake arithmetic), so rivals of LXT already run labelling at large scale. Date: EDGAR filing date."}]}},"evaluation":{"alternatives":{"level":1,"evidence":[{"url":"https://techcrunch.com/2026/06/29/arena-the-ai-leaderboard-everyone-uses-is-now-a-100m-business/","date":"2026-06-29","support":"TechCrunch says human-labelling firms Mercor, Surge and Scale AI compete with Arena \"for the same dollar\" in evaluation (paragraph 7); LXT's evaluation work has several qualified substitutes. Full date."},{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02-26","support":"Innodata's FY2025 10-K (Item 1, Competition) names Appen, CloudFactory, Surge AI, Invisible, Turing, Mercor, TELUS Digital and Scale AI as AI-data competitors; LXT work has many qualified homes. Date: EDGAR filing date."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02-26","support":"Innodata's 10-K (Item 1A) says customer agreements are generally terminable on about 30-90 days' notice, including on selecting another provider; LXT volume can be moved on that timescale. Date: EDGAR filing date."},{"url":"https://fortune.com/2025/06/19/openai-is-phasing-out-scale-ai-work-following-startups-meta-deal/","date":"2025-06-19","support":"OpenAI phased out Scale AI work within days of the Meta deal and relied increasingly on other providers such as Mercor (paragraphs 1, 9); LXT's expert-data work is similarly portable. Full date."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://techcrunch.com/2026/06/29/arena-the-ai-leaderboard-everyone-uses-is-now-a-100m-business/","date":"2026-06-29","support":"TechCrunch says human-labelling firms Mercor, Surge and Scale AI compete with Arena \"for the same dollar\" in evaluation (paragraph 7); LXT's evaluation work has several qualified substitutes. Full date."}]}}}},"b":{"schema":"5lc-index-rating/1","key":"models-data-evaluation","entityId":"lxt","rater":"b","model":"claude-sonnet-5-5","ratedAt":"2026-10-22","functions":{"labelling-multimodal":{"alternatives":{"level":1,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02","support":"Competition section names nine AI-data specialists (including Appen, CloudFactory, Surge, Invisible, Turing, Mercor, TELUS Digital and Scale) plus large services firms; LXT is one of many. Date: month only, taken from the 20 Feb 2026 cover-page share count for the FY2025 filing."},{"url":"https://www.lxt.ai/","date":"undated","support":"LXT lists 10M+ contributors and 250K+ domain experts across data collection and annotation (page undated; the footer shows only a 2026 copyright)."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://announcements.asx.com.au/asxpdf/20250226/pdf/06fyjdglv66466.pdf","date":"2025-02-26","support":"Google's termination of its Appen contract in Q1 FY24 cut Appen's FY24 group revenue 14.2% to $234.3M, showing a large buyer can withdraw a labelling programme within a quarter (ASX, 26 Feb 2025)."},{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02","support":"Innodata says its customer agreements are generally terminable on 30 to 90 days' notice and do not commit customers to future purchases (risk factors). Date: month only."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://www.investegate.co.uk/announcement/rns/rws-holdings--rws/half-year-financial-report/9612324","date":"2026-06-11","support":"RWS TrainAI revenue (Generate segment) grew 44.7% to GBP 99.4M in the half to March 2026, with a new global technology client won, so incumbents are adding programmes and volume has somewhere to go (11 Jun 2026)."},{"url":"https://www.sec.gov/Archives/edgar/data/0000868675/000110465926088975/tm2615221d2_ex99-1.htm","date":"2026-07-31","support":"TELUS Digital's AI and data solutions line is shrinking on client ramp-downs, which suggests spare capacity among incumbents (31 Jul 2026)."}]}},"evaluation":{"alternatives":{"level":1,"evidence":[{"url":"https://www.lxt.ai/","date":"undated","support":"LXT lists human-led evaluation and LLM red-teaming next to collection and annotation (page undated; the footer shows only a 2026 copyright)."},{"url":"https://labs.scale.com/leaderboard","date":"undated","support":"Scale Labs publishes 19 active leaderboards (page undated; the footer shows only a 2026 copyright)."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://mercor.com/blog/introducing-apex-ai-productivity-index/","date":"2025-10-01","support":"Mercor built its APEX v1 test set with about 100 experts in four professions (200 of 300 prompts approved), suggesting a comparable expert-written set is a matter of months (my inference, 1 Oct 2025)."},{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02","support":"Innodata says its customer agreements are generally terminable on 30 to 90 days' notice and do not commit customers to future purchases (risk factors). Date: month only."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://scale.com/blog/scales-next-era-building-for-2026","date":"2026-01-22","support":"Scale's lab says its lab introduced 15 new benchmarks and published over 450 evaluations of more than 50 models, a scale other vendors also reach (company blog, 22 Jan 2026)."},{"url":"https://welodata.ai/","date":"undated","support":"Welo Data lists evaluations, benchmarking, model selection and red teaming among its services (page undated; the footer shows only a 2026 copyright)."}]}}}}},"dataocean-ai":{"a":{"schema":"5lc-index-rating/1","key":"models-data-evaluation","entityId":"dataocean-ai","rater":"a","model":"claude-opus-5-5","ratedAt":"2026-10-22","functions":{"labelling-multimodal":{"alternatives":{"level":1,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02-26","support":"Innodata's FY2025 10-K (Item 1, Competition) names Appen, CloudFactory, Surge AI, Invisible, Turing, Mercor, TELUS Digital and Scale AI as AI-data competitors; DataOcean AI work has many qualified homes. Date: EDGAR filing date."},{"url":"https://www.sec.gov/Archives/edgar/data/1829864/000182986426000012/task-20251231.htm","date":"2026-03-05","support":"TaskUs's FY2025 10-K (Item 1, Our Competition) calls AI data services a \"large, rapidly changing, and fragmented global market\"; many vendors can take DataOcean AI's labelling work. Date: EDGAR filing date."},{"url":"https://jamestown.org/ai-data-annotation-sector-sees-misaligned-progress/","date":"2026-10-14","support":"China Brief says Chinese labelling is labour-intensive and crowdsourcing remains dominant, with about 140,000 people employed at state labelling bases (sections 'Foundational Policies', 'Transition'); DataOcean AI's work is widely substitutable. Full date."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://www.searchenginejournal.com/google-terminates-contract-with-appen-for-search-quality-raters/506006/","date":"2024-01-21","support":"Google gave Appen notice on 20 January 2024 and ended its services by 19 March 2024 (paragraphs 1-2), a two-month exit; DataOcean AI's volume could move similarly. Full date."},{"url":"https://www.sec.gov/Archives/edgar/data/1829864/000182986426000012/task-20251231.htm","date":"2026-03-05","support":"TaskUs's 10-K (Item 1A) says clients can terminate for convenience or move work in-house or to competitors; DataOcean AI's programmes are portable within weeks to months. Date: EDGAR filing date."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/1829864/000182986426000012/task-20251231.htm","date":"2026-03-05","support":"TaskUs's 10-K (Item 1) puts AI Services at 18% of US$1,183.5m 2025 revenue (about US$213m, 5LayerCake arithmetic), so rivals of DataOcean AI already run labelling at large scale. Date: EDGAR filing date."}]}}}},"b":{"schema":"5lc-index-rating/1","key":"models-data-evaluation","entityId":"dataocean-ai","rater":"b","model":"claude-sonnet-5-5","ratedAt":"2026-10-22","functions":{"labelling-multimodal":{"alternatives":{"level":2,"evidence":[{"url":"https://finance.sina.cn/stock/jdts/2026-10-09/detail-iniurwaf7778162.d.html","date":"2026-10-09","support":"DataOcean supplies Zhipu, Meta, Alibaba, Tencent, Moonshot and MiniMax with Chinese-language speech and vision data (H1 2026 results, 9 Oct 2026)."},{"url":"https://announcements.asx.com.au/asxpdf/20260827/pdf/0739d4q7lyh3w2.pdf","date":"2026-08-27","support":"Appen China (annualised run-rate over $175M, June 2026) sells to Chinese model builders; with Datatang that is the extent of the Chinese-language suppliers I verified (others may exist; evidence thin)."},{"url":"https://www.datatang.com/","date":"undated","support":"Datatang sells licensed datasets, custom collection and annotation for speech, vision and LLMs (page undated; the footer shows only a 2026 copyright)."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://announcements.asx.com.au/asxpdf/20250226/pdf/06fyjdglv66466.pdf","date":"2025-02-26","support":"Google's termination of its Appen contract in Q1 FY24 cut Appen's FY24 group revenue 14.2% to $234.3M, showing a large buyer can withdraw a labelling programme within a quarter (ASX, 26 Feb 2025)."},{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02","support":"Innodata says its customer agreements are generally terminable on 30 to 90 days' notice and do not commit customers to future purchases (risk factors). Date: month only."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://finance.sina.cn/stock/jdts/2026-10-09/detail-iniurwaf7778162.d.html","date":"2026-10-09","support":"DataOcean's H1 2026 revenue was RMB 192M (+22%); Appen China alone runs above $175M annualised, so no capacity shortfall in alternatives (9 Oct 2026)."},{"url":"https://announcements.asx.com.au/asxpdf/20260827/pdf/0739d4q7lyh3w2.pdf","date":"2026-08-27","support":"Appen China revenue rose 80% to $76.2M in H1 FY26 (27 Aug 2026)."}]}}}}},"encord":{"a":{"schema":"5lc-index-rating/1","key":"models-data-evaluation","entityId":"encord","rater":"a","model":"claude-opus-5-5","ratedAt":"2026-10-22","functions":{"data-tooling":{"alternatives":{"level":1,"evidence":[{"url":"https://lightly.ai/blog/the-10-best-encord-alternatives-in-2026-a-practical-guide-for-ml-team","date":"2026-05-18","support":"A May 2026 vendor guide lists ten labelling platforms (Labelbox, SuperAnnotate, Scale AI, V7, CVAT, Label Studio, Roboflow, Kili, Voxel51, Lightly) plus Dataloop as alternatives; many substitutes for Encord. Full date; author is a rival vendor."},{"url":"https://github.com/HumanSignal/label-studio","date":"undated","support":"Label Studio is an open-source, multi-type data labelling tool (text, audio, images, video, time series) with about 28.5k GitHub stars; a free substitute for Encord's software. Page undated."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://aws.amazon.com/about-aws/whats-new/2026/06/aws-service-availability/","date":"2026-06","support":"AWS ended SageMaker Ground Truth Plus on 30 June 2026 and closed Ground Truth to new customers from 30 July 2026: labelling-software users are expected to migrate within weeks. Relevant to Encord. Month taken from the URL path."},{"url":"https://docs.supervisely.com/import-and-export/migrations/migration-labelbox","date":"undated","support":"Supervisely offers a tool that copies multiple Labelbox projects into its platform; ready migration tooling keeps a switch away from Encord-type platforms to weeks. Page undated."},{"url":"https://lightly.ai/blog/the-10-best-encord-alternatives-in-2026-a-practical-guide-for-ml-team","date":"2026-05-18","support":"The guide advises \"Run the top two on a representative slice before you commit\", a short trial-based switch for platforms like Encord. Full date; author is a rival vendor."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://github.com/HumanSignal/label-studio","date":"undated","support":"Label Studio is an open-source, multi-type data labelling tool (text, audio, images, video, time series) with about 28.5k GitHub stars; a free substitute for Encord's software. Page undated."},{"url":"https://lightly.ai/blog/the-10-best-encord-alternatives-in-2026-a-practical-guide-for-ml-team","date":"2026-05-18","support":"A May 2026 vendor guide lists ten labelling platforms (Labelbox, SuperAnnotate, Scale AI, V7, CVAT, Label Studio, Roboflow, Kili, Voxel51, Lightly) plus Dataloop as alternatives; many substitutes for Encord. Full date; author is a rival vendor."}]}}}},"b":{"schema":"5lc-index-rating/1","key":"models-data-evaluation","entityId":"encord","rater":"b","model":"claude-sonnet-5-5","ratedAt":"2026-10-22","functions":{"data-tooling":{"alternatives":{"level":1,"evidence":[{"url":"https://github.com/HumanSignal/label-studio","date":"undated","support":"Label Studio is an Apache-2.0 open-source labelling tool with about 28.5k GitHub stars covering images, audio, text, video and time series (page undated; the footer shows only a 2026 copyright)."},{"url":"https://github.com/cvat-ai/cvat","date":"undated","support":"CVAT is an MIT-licensed annotation platform for image, video, audio and 3D data with about 17k GitHub stars (page undated; the footer shows only a 2026 copyright)."},{"url":"https://pulse2.com/encord-60-million-series-c-raised-to-scale-ai-native-data-infrastructure","date":"2026-02-27","support":"Encord sells a commercial multimodal annotation and curation platform to 300+ AI teams (press release, 27 Feb 2026)."},{"url":"https://www.vcaonline.com/news/2025071524/superannotate-raises-13-5-million-in-new-funding-from-dell-technologies-capital/","date":"2025-07-15","support":"SuperAnnotate sells a commercial annotation and evaluation platform to brands such as Databricks, Canva and Motorola Solutions (15 Jul 2025, company claim)."}]},"switchingTime":{"level":2,"evidence":[{"url":"https://pulse2.com/encord-60-million-series-c-raised-to-scale-ai-native-data-infrastructure","date":"2026-02-27","support":"Encord says hosted data grew from about 1 PB to over 5 PB in 12 months across 300+ AI teams, so migrating a team's multimodal data plausibly takes months (my inference; 27 Feb 2026, company statement)."},{"url":"https://github.com/HumanSignal/label-studio","date":"undated","support":"Open-source tools can be self-hosted and imported to without licence or procurement, shortening any switch (page undated; the footer shows only a 2026 copyright)."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://github.com/cvat-ai/cvat","date":"undated","support":"CVAT's MIT-licensed core already covers bounding boxes, polygons, masks, keypoints, cuboids and 3D point clouds (page undated; the footer shows only a 2026 copyright)."},{"url":"https://github.com/HumanSignal/label-studio","date":"undated","support":"Label Studio covers images, audio, text, video and time series with Apache-2.0 licensing (page undated; the footer shows only a 2026 copyright)."}]}}}}},"datatang":{"a":{"schema":"5lc-index-rating/1","key":"models-data-evaluation","entityId":"datatang","rater":"a","model":"claude-opus-5-5","ratedAt":"2026-10-22","functions":{"labelling-multimodal":{"alternatives":{"level":1,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02-26","support":"Innodata's FY2025 10-K (Item 1, Competition) names Appen, CloudFactory, Surge AI, Invisible, Turing, Mercor, TELUS Digital and Scale AI as AI-data competitors; Datatang work has many qualified homes. Date: EDGAR filing date."},{"url":"https://www.sec.gov/Archives/edgar/data/1829864/000182986426000012/task-20251231.htm","date":"2026-03-05","support":"TaskUs's FY2025 10-K (Item 1, Our Competition) calls AI data services a \"large, rapidly changing, and fragmented global market\"; many vendors can take Datatang's labelling work. Date: EDGAR filing date."},{"url":"https://jamestown.org/ai-data-annotation-sector-sees-misaligned-progress/","date":"2026-10-14","support":"China Brief says Chinese labelling is labour-intensive and crowdsourcing remains dominant, with about 140,000 people employed at state labelling bases (sections 'Foundational Policies', 'Transition'); Datatang's work is widely substitutable. Full date."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://www.searchenginejournal.com/google-terminates-contract-with-appen-for-search-quality-raters/506006/","date":"2024-01-21","support":"Google gave Appen notice on 20 January 2024 and ended its services by 19 March 2024 (paragraphs 1-2), a two-month exit; Datatang's volume could move similarly. Full date."},{"url":"https://www.sec.gov/Archives/edgar/data/1829864/000182986426000012/task-20251231.htm","date":"2026-03-05","support":"TaskUs's 10-K (Item 1A) says clients can terminate for convenience or move work in-house or to competitors; Datatang's programmes are portable within weeks to months. Date: EDGAR filing date."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/1829864/000182986426000012/task-20251231.htm","date":"2026-03-05","support":"TaskUs's 10-K (Item 1) puts AI Services at 18% of US$1,183.5m 2025 revenue (about US$213m, 5LayerCake arithmetic), so rivals of Datatang already run labelling at large scale. Date: EDGAR filing date."}]}}}},"b":{"schema":"5lc-index-rating/1","key":"models-data-evaluation","entityId":"datatang","rater":"b","model":"claude-sonnet-5-5","ratedAt":"2026-10-22","functions":{"labelling-multimodal":{"alternatives":{"level":2,"evidence":[{"url":"https://www.datatang.com/","date":"undated","support":"Datatang (founded 2010) sells licensed datasets, custom collection and an annotation platform, mainly to Chinese model builders (page undated; the footer shows only a 2026 copyright)."},{"url":"https://announcements.asx.com.au/asxpdf/20260827/pdf/0739d4q7lyh3w2.pdf","date":"2026-08-27","support":"Appen China (over $175M annualised, June 2026) sells to Chinese model builders; with DataOcean (Zhipu, Alibaba, Tencent customers) that is the extent of the Chinese-language suppliers I verified (evidence thin)."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://announcements.asx.com.au/asxpdf/20250226/pdf/06fyjdglv66466.pdf","date":"2025-02-26","support":"Google's termination of its Appen contract in Q1 FY24 cut Appen's FY24 group revenue 14.2% to $234.3M, showing a large buyer can withdraw a labelling programme within a quarter (ASX, 26 Feb 2025)."},{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02","support":"Innodata says its customer agreements are generally terminable on 30 to 90 days' notice and do not commit customers to future purchases (risk factors). Date: month only."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://www.datatang.com/","date":"undated","support":"Datatang's homepage gives conflicting customer counts (1,000+ versus 5,000+) and no revenue (page undated; the footer shows only a 2026 copyright); nothing shows peers lack capacity."},{"url":"https://www.investegate.co.uk/announcement/rns/rws-holdings--rws/half-year-financial-report/9612324","date":"2026-06-11","support":"RWS TrainAI revenue (Generate segment) grew 44.7% to GBP 99.4M in the half to March 2026, with a new global technology client won, so incumbents are adding programmes and volume has somewhere to go (11 Jun 2026)."}]}}}}},"sama":{"a":{"schema":"5lc-index-rating/1","key":"models-data-evaluation","entityId":"sama","rater":"a","model":"claude-opus-5-5","ratedAt":"2026-10-22","functions":{"labelling-multimodal":{"alternatives":{"level":1,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02-26","support":"Innodata's FY2025 10-K (Item 1, Competition) names Appen, CloudFactory, Surge AI, Invisible, Turing, Mercor, TELUS Digital and Scale AI as AI-data competitors; Sama work has many qualified homes. Date: EDGAR filing date."},{"url":"https://www.sec.gov/Archives/edgar/data/1829864/000182986426000012/task-20251231.htm","date":"2026-03-05","support":"TaskUs's FY2025 10-K (Item 1, Our Competition) calls AI data services a \"large, rapidly changing, and fragmented global market\"; many vendors can take Sama's labelling work. Date: EDGAR filing date."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://www.searchenginejournal.com/google-terminates-contract-with-appen-for-search-quality-raters/506006/","date":"2024-01-21","support":"Google gave Appen notice on 20 January 2024 and ended its services by 19 March 2024 (paragraphs 1-2), a two-month exit; Sama's volume could move similarly. Full date."},{"url":"https://www.sec.gov/Archives/edgar/data/1829864/000182986426000012/task-20251231.htm","date":"2026-03-05","support":"TaskUs's 10-K (Item 1A) says clients can terminate for convenience or move work in-house or to competitors; Sama's programmes are portable within weeks to months. Date: EDGAR filing date."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/1829864/000182986426000012/task-20251231.htm","date":"2026-03-05","support":"TaskUs's 10-K (Item 1) puts AI Services at 18% of US$1,183.5m 2025 revenue (about US$213m, 5LayerCake arithmetic), so rivals of Sama already run labelling at large scale. Date: EDGAR filing date."}]}}}},"b":{"schema":"5lc-index-rating/1","key":"models-data-evaluation","entityId":"sama","rater":"b","model":"claude-sonnet-5-5","ratedAt":"2026-10-22","functions":{"labelling-multimodal":{"alternatives":{"level":1,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02","support":"Competition section names nine AI-data specialists (including Appen, CloudFactory, Surge, Invisible, Turing, Mercor, TELUS Digital and Scale) plus large services firms; Sama is one of many. Date: month only, taken from the 20 Feb 2026 cover-page share count for the FY2025 filing."},{"url":"https://www.sec.gov/Archives/edgar/data/0001829864/000182986426000147/earningsreleaseex991q22026.htm","date":"2026-08-05","support":"TaskUs reports AI Services as its fastest-growing line, up 25.8% year over year in Q2 2026, a large listed provider of AI data work (release of 5 Aug 2026)."},{"url":"https://announcements.asx.com.au/asxpdf/20260827/pdf/0739d4q7lyh3w2.pdf","date":"2026-08-27","support":"Appen China's annualised run-rate passed $175M in June 2026 and Appen Global is rebuilding; Appen is a further large vendor (ASX, 27 Aug 2026)."}]},"switchingTime":{"level":2,"evidence":[{"url":"https://www.sama.com/","date":"undated","support":"Sama says its customers stay 8 years on average (company claim) and cites a Walmart programme over 2.5M+ SKUs, hinting at embedded, multi-month onboarding (page undated; the footer shows only a 2026 copyright)."},{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02","support":"Innodata says its customer agreements are generally terminable on 30 to 90 days' notice and do not commit customers to future purchases (risk factors). Date: month only."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://www.investegate.co.uk/announcement/rns/rws-holdings--rws/half-year-financial-report/9612324","date":"2026-06-11","support":"RWS TrainAI revenue (Generate segment) grew 44.7% to GBP 99.4M in the half to March 2026, with a new global technology client won, so incumbents are adding programmes and volume has somewhere to go (11 Jun 2026)."},{"url":"https://www.sec.gov/Archives/edgar/data/0000868675/000110465926088975/tm2615221d2_ex99-1.htm","date":"2026-07-31","support":"TELUS Digital's AI and data solutions line is shrinking on client ramp-downs, which suggests spare capacity among incumbents (31 Jul 2026)."}]}}}}},"halluminate":{"a":{"schema":"5lc-index-rating/1","key":"models-data-evaluation","entityId":"halluminate","rater":"a","model":"claude-opus-5-5","ratedAt":"2026-10-22","functions":{"rl-environments":{"alternatives":{"level":1,"evidence":[{"url":"https://newsletter.semianalysis.com/p/rl-environments-and-rl-for-science","date":"2026-01-06","support":"SemiAnalysis counts \"more than 35 companies\" building RL environments, mostly seed-stage, each serving 1-3 customers (vendor section); Halluminate faces many substitutes. Full date."},{"url":"https://www.ycombinator.com/companies/halluminate","date":"undated","support":"Halluminate is a five-person YC Summer 2025 company that says it built environments with four of the five leading closed US labs; many similar builders exist. Page undated."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://techcrunch.com/2025/09/16/silicon-valley-bets-big-on-environments-to-train-ai-agents/","date":"2025-09-16","support":"TechCrunch says \"there's no shortage of startups\" supplying environments and that big labs also build them in-house (paragraphs 3-4); alternatives to Halluminate are plentiful. Full date."},{"url":"https://newsletter.semianalysis.com/p/rl-environments-and-rl-for-science","date":"2026-01-06","support":"SemiAnalysis says most environment vendors sell under exclusive contracts and UI-gym sites cost about US$20,000 each as one-time purchases (vendor and pricing sections); replacing Halluminate's pipeline means commissioning new builds. Full date."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://techcrunch.com/2026/09/22/snorkel-ai-triples-valuation-to-3-5b-as-demand-for-ai-training-data-booms/","date":"2026-09-22","support":"TechCrunch names Mercor (US$2bn gross), Handshake (US$1bn) and micro1 (US$500m) as scaled peers, and Snorkel at US$375m run-rate (company says) (paragraphs 4-5); several large alternatives to Halluminate. Full date."}]}}}},"b":{"schema":"5lc-index-rating/1","key":"models-data-evaluation","entityId":"halluminate","rater":"b","model":"claude-sonnet-5-5","ratedAt":"2026-10-22","functions":{"rl-environments":{"alternatives":{"level":1,"evidence":[{"url":"https://newsletter.semianalysis.com/p/rl-environments-and-rl-for-science","date":"2026-01-06","support":"More than 35 companies sell RL environments (mostly seed-stage) and Anthropic likely works with more than a dozen of them (SemiAnalysis, 6 Jan 2026)."},{"url":"https://www.halluminate.ai/","date":"undated","support":"Halluminate builds finance and consulting benchmarks and environments (an 88-problem diligence bench dated 11 Aug 2026) and names no customers (page undated; the footer shows only a 2026 copyright)."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://epoch.ai/gradient-updates/state-of-rl-envs","date":"2026-01-12","support":"Environment contracts are often six to seven figures per quarter and can be non-exclusive, so labs can add or swap vendors quarter by quarter; Epoch also reports 'substantially more' in-housing (12 Jan 2026)."},{"url":"https://newsletter.semianalysis.com/p/rl-environments-and-rl-for-science","date":"2026-01-06","support":"Anthropic is 'often the first customer' of new environment startups and wants a broad vendor ecosystem so the product is commoditised (SemiAnalysis, 6 Jan 2026)."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://epoch.ai/gradient-updates/state-of-rl-envs","date":"2026-01-12","support":"Quality at scale is the main bottleneck for all vendors; interviewees rank robustness to reward hacking first and say domain knowledge matters more than ML skill (12 Jan 2026). No vendor is described as uniquely capable."},{"url":"https://newsletter.semianalysis.com/p/rl-environments-and-rl-for-science","date":"2026-01-06","support":"Reports 'significant diversity in quality or focus' among vendors, with Turing on UI gyms and others on software engineering, rather than a single leader (6 Jan 2026)."}]}}}}},"gray-swan":{"a":{"schema":"5lc-index-rating/1","key":"models-data-evaluation","entityId":"gray-swan","rater":"a","model":"claude-opus-5-5","ratedAt":"2026-10-22","functions":{"evaluation":{"alternatives":{"level":2,"evidence":[{"url":"https://arxiv.org/html/2412.16720v1","date":"2024-12-21","support":"OpenAI's o1 system card lists external testers Gray Swan, Apollo Research, METR, US and UK AISI, and its own Red Teaming Network (sections 4, 4.4): several red-teaming alternatives exist. arXiv v1 submission date."},{"url":"https://www.grayswan.ai/news/gray-swan-announces-series-a","date":"2026-05-28","support":"Gray Swan says it is cited in 11 recent frontier system cards (paragraph 3); other red-teamers appear in the same cards. Full date."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://arxiv.org/html/2412.16720v1","date":"2024-12-21","support":"Red-teaming engagements are campaign-based (Gray Swan Arena jailbreak challenge, section 4.4.2); a lab can commission another tester or its own network within weeks. arXiv v1 submission date."}]},"capabilityGap":{"level":2,"evidence":[{"url":"https://www.grayswan.ai/news/gray-swan-announces-series-a","date":"2026-05-28","support":"Gray Swan says its Arena has 15,000+ participants and 1M+ real-world attack trajectories (paragraph 13), a community scale rivals do not report: a small gap. Full date."}]}}}},"b":{"schema":"5lc-index-rating/1","key":"models-data-evaluation","entityId":"gray-swan","rater":"b","model":"claude-sonnet-5-5","ratedAt":"2026-10-22","functions":{"evaluation":{"alternatives":{"level":2,"evidence":[{"url":"https://grayswan.ai/solutions/for-model-builders/adversarial-evaluation","date":"undated","support":"Gray Swan claims 15,000+ red teamers and 'more external evaluation citations' than any other red-teaming partner (page undated; the footer shows only a 2026 copyright)."},{"url":"https://nebius.com/newsroom/nebius-welcomes-bezos-expeditions-as-lead-investor-in-ai-data-business-toloka","date":"2025-05-07","support":"Toloka also sells agent safety evaluation and red teaming to labs (7 May 2025); LXT and Welo list red teaming too, giving four to five vendors."}]},"switchingTime":{"level":2,"evidence":[{"url":"https://grayswan.ai/solutions/for-model-builders/adversarial-evaluation","date":"undated","support":"Gray Swan sells pre-release adversarial evaluation to model builders (page undated; the footer shows only a 2026 copyright); no source gives a switching time, so 3-12 months is my judgement for contracting and onboarding another red-team provider. Evidence thin."}]},"capabilityGap":{"level":2,"evidence":[{"url":"https://grayswan.ai/solutions/for-model-builders/adversarial-evaluation","date":"undated","support":"Its differentiator is a standing red-team community of 15,000+ attackers (company claim) (page undated; the footer shows only a 2026 copyright); rivals do not publish network sizes, so the gap is judged small. Evidence thin."}]}}}}},"superannotate":{"a":{"schema":"5lc-index-rating/1","key":"models-data-evaluation","entityId":"superannotate","rater":"a","model":"claude-opus-5-5","ratedAt":"2026-10-22","functions":{"data-tooling":{"alternatives":{"level":1,"evidence":[{"url":"https://lightly.ai/blog/the-10-best-encord-alternatives-in-2026-a-practical-guide-for-ml-team","date":"2026-05-18","support":"A May 2026 vendor guide lists ten labelling platforms (Labelbox, SuperAnnotate, Scale AI, V7, CVAT, Label Studio, Roboflow, Kili, Voxel51, Lightly) plus Dataloop as alternatives; many substitutes for SuperAnnotate. Full date; author is a rival vendor."},{"url":"https://github.com/HumanSignal/label-studio","date":"undated","support":"Label Studio is an open-source, multi-type data labelling tool (text, audio, images, video, time series) with about 28.5k GitHub stars; a free substitute for SuperAnnotate's software. Page undated."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://aws.amazon.com/about-aws/whats-new/2026/06/aws-service-availability/","date":"2026-06","support":"AWS ended SageMaker Ground Truth Plus on 30 June 2026 and closed Ground Truth to new customers from 30 July 2026: labelling-software users are expected to migrate within weeks. Relevant to SuperAnnotate. Month taken from the URL path."},{"url":"https://docs.supervisely.com/import-and-export/migrations/migration-labelbox","date":"undated","support":"Supervisely offers a tool that copies multiple Labelbox projects into its platform; ready migration tooling keeps a switch away from SuperAnnotate-type platforms to weeks. Page undated."},{"url":"https://lightly.ai/blog/the-10-best-encord-alternatives-in-2026-a-practical-guide-for-ml-team","date":"2026-05-18","support":"The guide advises \"Run the top two on a representative slice before you commit\", a short trial-based switch for platforms like SuperAnnotate. Full date; author is a rival vendor."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://github.com/HumanSignal/label-studio","date":"undated","support":"Label Studio is an open-source, multi-type data labelling tool (text, audio, images, video, time series) with about 28.5k GitHub stars; a free substitute for SuperAnnotate's software. Page undated."},{"url":"https://lightly.ai/blog/the-10-best-encord-alternatives-in-2026-a-practical-guide-for-ml-team","date":"2026-05-18","support":"A May 2026 vendor guide lists ten labelling platforms (Labelbox, SuperAnnotate, Scale AI, V7, CVAT, Label Studio, Roboflow, Kili, Voxel51, Lightly) plus Dataloop as alternatives; many substitutes for SuperAnnotate. Full date; author is a rival vendor."}]}}}},"b":{"schema":"5lc-index-rating/1","key":"models-data-evaluation","entityId":"superannotate","rater":"b","model":"claude-sonnet-5-5","ratedAt":"2026-10-22","functions":{"data-tooling":{"alternatives":{"level":1,"evidence":[{"url":"https://github.com/HumanSignal/label-studio","date":"undated","support":"Label Studio is an Apache-2.0 open-source labelling tool with about 28.5k GitHub stars covering images, audio, text, video and time series (page undated; the footer shows only a 2026 copyright)."},{"url":"https://github.com/cvat-ai/cvat","date":"undated","support":"CVAT is an MIT-licensed annotation platform for image, video, audio and 3D data with about 17k GitHub stars (page undated; the footer shows only a 2026 copyright)."},{"url":"https://pulse2.com/encord-60-million-series-c-raised-to-scale-ai-native-data-infrastructure","date":"2026-02-27","support":"Encord sells a commercial multimodal annotation and curation platform to 300+ AI teams (press release, 27 Feb 2026)."},{"url":"https://www.vcaonline.com/news/2025071524/superannotate-raises-13-5-million-in-new-funding-from-dell-technologies-capital/","date":"2025-07-15","support":"SuperAnnotate sells a commercial annotation and evaluation platform to brands such as Databricks, Canva and Motorola Solutions (15 Jul 2025, company claim)."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://www.vcaonline.com/news/2025071524/superannotate-raises-13-5-million-in-new-funding-from-dell-technologies-capital/","date":"2025-07-15","support":"SuperAnnotate says its software revenue grew 5x in 2024 with customers such as Databricks and Canva, i.e. buyers adopt a platform quickly (15 Jul 2025, company claim; evidence thin)."},{"url":"https://github.com/HumanSignal/label-studio","date":"undated","support":"Open-source tools can be self-hosted and imported to without licence or procurement, shortening any switch (page undated; the footer shows only a 2026 copyright)."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://github.com/cvat-ai/cvat","date":"undated","support":"CVAT's MIT-licensed core already covers bounding boxes, polygons, masks, keypoints, cuboids and 3D point clouds (page undated; the footer shows only a 2026 copyright)."},{"url":"https://github.com/HumanSignal/label-studio","date":"undated","support":"Label Studio covers images, audio, text, video and time series with Apache-2.0 licensing (page undated; the footer shows only a 2026 copyright)."}]}}}}},"irregular":{"a":{"schema":"5lc-index-rating/1","key":"models-data-evaluation","entityId":"irregular","rater":"a","model":"claude-opus-5-5","ratedAt":"2026-10-22","functions":{"evaluation":{"alternatives":{"level":3,"evidence":[{"url":"https://www.securityweek.com/irregular-raises-80-million-for-ai-security-testing-lab/","date":"2025-09-17","support":"Irregular says it works side by side with OpenAI, Google and Anthropic on frontier-model security evaluation (paragraph 5); few commercial labs offer this cyber-offence testing. Full date."},{"url":"https://arxiv.org/html/2412.16720v1","date":"2024-12-21","support":"The o1 card's external testers (Apollo, METR, Gray Swan, US/UK AISI) cover scheming, agentic and jailbreak risks, not cyber-offence ranges (sections 4, 4.4); few alternatives, thin evidence. arXiv v1 submission date."}]},"switchingTime":{"level":2,"evidence":[{"url":"https://www.securityweek.com/irregular-raises-80-million-for-ai-security-testing-lab/","date":"2025-09-17","support":"Irregular runs controlled simulations of models on real-world cyber threats (paragraphs 1-5); a replacement would need its own ranges and pre-release access, likely 3-12 months. Full date; inference."}]},"capabilityGap":{"level":2,"evidence":[{"url":"https://arxiv.org/html/2412.16720v1","date":"2024-12-21","support":"OpenAI ran its own cyber capability tests and used government institutes for pre-deployment evaluation (section 4 introduction): in-house and AISI alternatives exist, a small gap. arXiv v1 submission date."}]}}}},"b":{"schema":"5lc-index-rating/1","key":"models-data-evaluation","entityId":"irregular","rater":"b","model":"claude-sonnet-5-5","ratedAt":"2026-10-22","functions":{"evaluation":{"alternatives":{"level":3,"evidence":[{"url":"https://techcrunch.com/2025/09/17/irregular-raises-80-million-to-secure-frontier-ai-models","date":"2025-09-17","support":"Irregular's work is cited in security evaluations for Claude 3.7 Sonnet and OpenAI's o3 and o4-mini, and its SOLVE framework is widely used (17 Sep 2025)."},{"url":"https://metr.org/","date":"undated","support":"METR says it has partnered with OpenAI, Anthropic, Google DeepMind, Meta and Amazon on frontier risk assessments, mainly autonomy and AI R&D capability rather than cyber (page undated; the footer shows only a 2026 copyright)."},{"url":"https://www.apolloresearch.ai/","date":"undated","support":"Apollo Research runs pre-deployment evaluations for strategic deception and scheming, also not cyber (page undated; the footer shows only a 2026 copyright); cyber-capability evaluators are therefore fewer (judgement, evidence thin)."}]},"switchingTime":{"level":2,"evidence":[{"url":"https://techcrunch.com/2025/09/17/irregular-raises-80-million-to-secure-frontier-ai-models","date":"2025-09-17","support":"Irregular runs simulated attacker-defender environments on pre-release models (17 Sep 2025); no source gives a replacement time, judged 3-12 months to build or contract an equivalent harness. Evidence thin."}]},"capabilityGap":{"level":3,"evidence":[{"url":"https://techcrunch.com/2025/09/17/irregular-raises-80-million-to-secure-frontier-ai-models","date":"2025-09-17","support":"SOLVE is described as 'widely used in the industry' and Irregular raised $80M at a reported $450M valuation to extend it (17 Sep 2025); no peer with a comparable cyber harness is named."},{"url":"https://metr.org/","date":"undated","support":"METR does not accept compensation for its lab evaluations and works on other risk domains (page undated; the footer shows only a 2026 copyright)."}]}}}}},"cloudfactory":{"a":{"schema":"5lc-index-rating/1","key":"models-data-evaluation","entityId":"cloudfactory","rater":"a","model":"claude-opus-5-5","ratedAt":"2026-10-22","functions":{"labelling-multimodal":{"alternatives":{"level":1,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02-26","support":"Innodata's FY2025 10-K (Item 1, Competition) names Appen, CloudFactory, Surge AI, Invisible, Turing, Mercor, TELUS Digital and Scale AI as AI-data competitors; CloudFactory work has many qualified homes. Date: EDGAR filing date."},{"url":"https://www.sec.gov/Archives/edgar/data/1829864/000182986426000012/task-20251231.htm","date":"2026-03-05","support":"TaskUs's FY2025 10-K (Item 1, Our Competition) calls AI data services a \"large, rapidly changing, and fragmented global market\"; many vendors can take CloudFactory's labelling work. Date: EDGAR filing date."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://www.searchenginejournal.com/google-terminates-contract-with-appen-for-search-quality-raters/506006/","date":"2024-01-21","support":"Google gave Appen notice on 20 January 2024 and ended its services by 19 March 2024 (paragraphs 1-2), a two-month exit; CloudFactory's volume could move similarly. Full date."},{"url":"https://www.sec.gov/Archives/edgar/data/1829864/000182986426000012/task-20251231.htm","date":"2026-03-05","support":"TaskUs's 10-K (Item 1A) says clients can terminate for convenience or move work in-house or to competitors; CloudFactory's programmes are portable within weeks to months. Date: EDGAR filing date."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/1829864/000182986426000012/task-20251231.htm","date":"2026-03-05","support":"TaskUs's 10-K (Item 1) puts AI Services at 18% of US$1,183.5m 2025 revenue (about US$213m, 5LayerCake arithmetic), so rivals of CloudFactory already run labelling at large scale. Date: EDGAR filing date."}]}}}},"b":{"schema":"5lc-index-rating/1","key":"models-data-evaluation","entityId":"cloudfactory","rater":"b","model":"claude-sonnet-5-5","ratedAt":"2026-10-22","functions":{"labelling-multimodal":{"alternatives":{"level":1,"evidence":[{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02","support":"Competition section names nine AI-data specialists (including Appen, CloudFactory, Surge, Invisible, Turing, Mercor, TELUS Digital and Scale) plus large services firms; CloudFactory is one of many. Date: month only, taken from the 20 Feb 2026 cover-page share count for the FY2025 filing."},{"url":"https://www.sec.gov/Archives/edgar/data/0001829864/000182986426000147/earningsreleaseex991q22026.htm","date":"2026-08-05","support":"TaskUs reports AI Services as its fastest-growing line, up 25.8% year over year in Q2 2026, a large listed provider of AI data work (release of 5 Aug 2026)."},{"url":"https://announcements.asx.com.au/asxpdf/20260827/pdf/0739d4q7lyh3w2.pdf","date":"2026-08-27","support":"Appen China's annualised run-rate passed $175M in June 2026 and Appen Global is rebuilding; Appen is a further large vendor (ASX, 27 Aug 2026)."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://announcements.asx.com.au/asxpdf/20250226/pdf/06fyjdglv66466.pdf","date":"2025-02-26","support":"Google's termination of its Appen contract in Q1 FY24 cut Appen's FY24 group revenue 14.2% to $234.3M, showing a large buyer can withdraw a labelling programme within a quarter (ASX, 26 Feb 2025)."},{"url":"https://www.sec.gov/Archives/edgar/data/903651/000110465926020655/inod-20251231x10k.htm","date":"2026-02","support":"Innodata says its customer agreements are generally terminable on 30 to 90 days' notice and do not commit customers to future purchases (risk factors). Date: month only."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://www.cloudfactory.com/","date":"undated","support":"CloudFactory now presents itself as an AI platform plus consulting firm, not mainly an annotation vendor (page undated; the footer shows only a 2026 copyright), so its labelling capacity is not scarce."},{"url":"https://www.investegate.co.uk/announcement/rns/rws-holdings--rws/half-year-financial-report/9612324","date":"2026-06-11","support":"RWS TrainAI revenue (Generate segment) grew 44.7% to GBP 99.4M in the half to March 2026, with a new global technology client won, so incumbents are adding programmes and volume has somewhere to go (11 Jun 2026)."}]}}}}},"afterquery":{},"artificial-analysis":{"a":{"schema":"5lc-index-rating/1","key":"models-data-evaluation","entityId":"artificial-analysis","rater":"a","model":"claude-opus-5-5","ratedAt":"2026-10-22","functions":{"evaluation":{"alternatives":{"level":2,"evidence":[{"url":"https://www.latent.space/p/state-of-evals-lmarenas-17b-vision","date":"2026-01-06","support":"Arena's 'State of Evals' episode names Artificial Analysis and AI.xyz as comparators (timestamp 00:06:02); with Arena and Scale's leaderboards, several alternatives exist. Full date."},{"url":"https://news.bgov.com/legal-ops-and-tech/scale-unveils-new-ai-model-leaderboard-system-to-rival-lmarena","date":"2025-09-22","support":"Scale publishes public model leaderboards (paragraphs 1-2), another independent ranking source. Full date."}]},"switchingTime":{"level":1,"evidence":[{"url":"https://artificialanalysis.ai/about","date":"undated","support":"Artificial Analysis describes itself as 'the independent benchmarking company for AI' with published leaderboards; users can move to other public rankings within weeks. Page undated (footer 2026)."}]},"capabilityGap":{"level":1,"evidence":[{"url":"https://techcrunch.com/2026/06/29/arena-the-ai-leaderboard-everyone-uses-is-now-a-100m-business/","date":"2026-06-29","support":"Arena's US$100m run-rate evaluation business (paragraph 1) is far larger than Artificial Analysis's share: no capability gap. Full date."}]}}}},"b":{"schema":"5lc-index-rating/1","key":"models-data-evaluation","entityId":"artificial-analysis","rater":"b","model":"claude-sonnet-5-5","ratedAt":"2026-10-22","functions":{"evaluation":{"alternatives":{"level":2,"evidence":[{"url":"https://artificialanalysis.ai/about","date":"undated","support":"Artificial Analysis benchmarks 500+ models, 100+ inference providers and 1,000+ endpoints (page undated)."},{"url":"https://labs.scale.com/leaderboard","date":"undated","support":"Scale Labs publishes 19 active leaderboards (page undated; the footer shows only a 2026 copyright); with Arena, Mercor's APEX and Prolific's HUMAINE I count four to five other independent evaluators."}]},"switchingTime":{"level":2,"evidence":[{"url":"https://artificialanalysis.ai/about","date":"undated","support":"Its coverage (1T+ evaluation tokens) is built from continuous benchmarking, so a rival needs months to match breadth; no source gives a timeline. Evidence thin; page undated."}]},"capabilityGap":{"level":3,"evidence":[{"url":"https://artificialanalysis.ai/about","date":"undated","support":"Artificial Analysis lists 500+ models, 100+ inference providers and 1,000+ endpoints benchmarked (page undated)."},{"url":"https://scale.com/blog/scales-next-era-building-for-2026","date":"2026-01-22","support":"By comparison Scale's lab says it published over 450 evaluations across more than 50 models (22 Jan 2026)."}]}}}}},"datacurve":{}},"validation":[]}