Training datasets
Companies that sell ready-made or custom training datasets.
Our coverage
Expert network Guidepoint launches Guidepoint Labs to sell expert data to AI labs
The unit ran in stealth for more than six months and says it already works with frontier labs. It draws on Guidepoint's network of more than 2 million advisors.
Humyn Labs raises $47 million to record real-world human data for robotics AI
Accel led $26 million in equity and Lighthouse Canton led $21 million in venture debt. The Bengaluru startup films people doing everyday tasks in more than 20 countries and sells the labeled footage to AI and robotics developers.
Scale AI's $12.5 million contractor settlement still shows no final approval
The California class covers Outlier and Remotasks contributors from Dec. 2020 to Feb. 2026. Up to $4.17 million can go to lawyers' fees.
Mercor posts $350 an hour for cardiologists to train AI
micro1 listed about 350 expert roles on Nov. 17. Outlier posts no hourly rate.
Mercor asks Texas court to send contractor misclassification suit to arbitration
Heidi White's suit says Mercor paid AI training experts as contractors on 1099s. Mercor's amended motion is pending.
Judge approves Google's $10 million purchase of Spirit Airlines data
Judge Sean Lane ruled from the bench on Nov. 17 after Google agreed to added privacy protections, Reuters reported. Mercor was the $7.5 million alternate bidder.
Companies (148), by published funding
Figures as published by each company or the source on its profile. n/a means no source found.
| Company | What it does | Based in | Funding | Partner program |
|---|---|---|---|---|
| Scale | Data-partnerships program for company data. Use the public program page, not outdated deal rumors. | United States | $15.9B | Yes |
| Applied Intuition | Simulation and autonomy software company serving automakers, defense buyers and other teams building machines that drive or fly themselves. | United States | $1.2B | n/a |
| Snorkel | Snorkel AI. Expert-built datasets, agent environments, benchmarks and evals, sold as products and custom services. | United States | $587M | n/a |
| Mercor | Expert-data vendor. Alternate bidder at $7.5 million in the Spirit Airlines de-identified data auction, per Business Insider. | United States | $483.6M | Yes |
| Handshake | Employer data-partnerships program; licenses professional documents, per the public page. | United States | $435.5M | Yes |
| Lightwheel | Sells physics-accurate simulation assets, demonstration data and validation tools to robotics teams. | United States | $293M | n/a |
| Alice | The former ActiveFence. Sells safety and security datasets, RL environments and red teaming to model labs, plus guardrail and moderation software. | Israel | $280M | n/a |
| TaskUs | Outsourcing firm running AI data operations: collection, labeling, RLHF, evals, red teaming and robotics data. | United States | $279M | n/a |
| Turing | Training data, evals and RL environments for coding, knowledge work and STEM, plus enterprise agent deployment. | United States | $225M | Yes |
| Labelbox | Labeling platform company that now sells RL environments, expert preference data and evals to frontier labs. | United States | $189M | No |
| Claru | Supplies custom and licensed video, robot demonstrations and annotation for robotics and world model work. | United States | $170M | n/a |
| Invisible Technologies | Supplies expert training data, fine-tuning, evals and custom RL environments, plus enterprise AI services. | United States | $144M | n/a |
| micro1 | Expert network and company-data program (Data Payout). Take rates, when quoted, come from its public program page. | United States | $135M+ | Yes |
| Mode | Public referral page for company-data licensing. The referral amount is stated on the live page. | United States | $120M+ in company-reported capital raised | Yes |
| Encord | Multimodal data platform with managed collection, labeling and eval services for vision and physical AI teams. | United States | $110M | n/a |
| Defined.ai | Data marketplace with a public partnership program. Partner-revenue figures on that page are company claims. | United States | $85M+ | No |
| Sama | Human-in-the-loop labeling, validation and evaluation services for vision and generative AI teams. | United States | $84.84M | n/a |
| David AI | Designs, records, labels and licenses human audio datasets and evaluates speech models. | United States | $80.5M | n/a |
| CloudFactory | Managed workforce and platform vendor for data collection, annotation, fine-tuning datasets and human feedback. | United Kingdom | $78M | n/a |
| Toloka | Human expert and synthetic data, agent environments, evals and red teaming for labs and developers. | Netherlands | $72M | n/a |
| Patronus AI | AI evaluation company that now builds simulated environments and training data, backed by research into where models fail. | United States | $70M | Yes |
| XDOF | Collects robot teleoperation data, runs real-world policy evals, and supplies robot systems and tooling. | United States | $70M | n/a |
| SuperAnnotate | AI data platform plus managed expert services for labeling, training data, evals and RL environments. | United States | $67.5M | No |
| Protege | Real-world data platform (healthcare and other modalities). a16z announced a $30 million round on January 8, 2026, and said the company already had a $25 million Series A. | United States | $65M | n/a |
| Centific | Provides multimodal training data, annotation, evals and simulated enterprise RL environments to labs and corporate AI teams. | United States | $60M | No |
| Mecka | Turns recorded human activity into robotics training data and offers data tools and evaluation support. | United States | $60M | n/a |
| microagi | Sells recordings of human manual work to labs and runs a service that trains and installs robots on factory lines. | Germany | $55M+ | n/a |
| Dataloop | Data platform plus on-demand labeling service for organizing, annotating and checking multimodal training data. | Israel | $50M | n/a |
| e2f | Renamed TrustScale in 2026. Sells custom datasets, expert labeling, evals and RL feedback through its ArgusRL platform. | United States | $50M | n/a |
| NewtonX | Vetted professional network that runs model evals, builds post-training data and audits labeling quality. | United States | $47M | n/a |
| Revelo | Developer talent firm selling code-focused training data, evals, benchmarks and RL environments. | United States | $46M | n/a |
| Contra Labs | Uses professional creatives to judge model output and to build benchmarks, feedback data and recorded design workflows. | United States | $44.5M | n/a |
| Halluminate | Sells finance-focused computer-use RL environments, simulated business apps, data and agent benchmarks. | United States | $38.5M | n/a |
| Prolific | Participant and expert panel platform for collecting human feedback, training data and evals. | United Kingdom | $33.4M | n/a |
| Centaur AI | Boston labeling company that scores competing human and AI annotators, sells evals, and licenses finished labeled datasets. | United States | $32.54M | n/a |
| Luel | Licenses multimodal datasets and runs paid contributor campaigns to collect rights-cleared audio, video and sensor data. | United States | $31.7M | n/a |
| AfterQuery | Sells AI labs and enterprises a mix of expert-made and synthetic datasets, RL environments and custom evals. | United States | $30.5M | Yes |
| Build AI | Films first-person video of workers in real factories and shops and sells it to robotics labs as stock or custom datasets. | United States | $30M+ | n/a |
| Moonlake AI | Turns text prompts or real-world captures into 3D simulation environments for robotics and autonomy research. | United States | $28M | n/a |
| Wirestock | Creator platform that licenses creator-made multimodal data to AI labs, as catalog sets or custom collections. | United States | $26M | n/a |
| Deccan AI | Supplies expert post-training data, human feedback, RL environments and evals, plus enterprise AI operations services. | United States | $25M | n/a |
| iMerit | Expert annotation and human feedback vendor that also sells the Ango Hub workflow platform. | United States | $24.5M | n/a |
| SimpleClosure | Wind-down firm whose AssetHub, per TechSpot on April 19, 2026 (citing Forbes), had facilitated nearly 100 data transactions with reported payouts of $10,000 to $100,000 per company. | United States | $20.5M | No |
| Taste Labs | Preference and reasoning data and eval environments for design quality, plus a Brand API for developers. | United States | $18.5M | n/a |
| Datacurve | Expert coding data vendor offering RL environments, benchmarks and evals to model labs and developer tool companies. | United States | $17.7M | n/a |
| Perle AI | Specialist-generated multimodal data, labeling, RLHF and evals through expert-led programs. | United States | $17.5M | n/a |
| HUD | Developer kit and cloud service for creating, running and scoring RL environments, plus a marketplace where vendors sell them. | United States | $16M | n/a |
| Preference Model | Builds RL environments that train models on machine learning research and engineering tasks. | United States | $16M | n/a |
| Sapien | Human-created and expert-reviewed data, labeling and output quality checks for AI developers. | Canada | $15.5M | n/a |
| Midcentury | Sells human action and gameplay datasets and cloud simulation tools to physical AI developers. | United States | $15M | n/a |
| Poseidon | Sells physical AI training data such as egocentric task video, robot manipulation, speech and motion capture. | per its site | $15M | n/a |
| Million | Research lab building front-end coding data, RL environments and evals, including the ReactBench benchmark. | per its Y Combinator profile and SEC filing | $14.15M over 3 rounds, per CB Insights | n/a |
| Kled | Marketplace that curates and licenses video, image, audio and text from contributors and rights holders. | United States | $14M | n/a |
| Replay | Pays companies for work history such as code, docs, tickets and CRM records, strips identifiers, and licenses the data. | United States | $12M+ | Yes |
| Chakra Labs | Writes expert task sets and graders, including computer-use environments, for model training and testing in finance, law and other fields. | United States | $10.1M | n/a |
| BioStack Platforms | Sells clinical and preclinical data, healthcare training environments and eval tools to medical and drug discovery AI teams. | United States | $10M | n/a |
| Collinear | Sells simulated environments, agent evals and synthetic training data to labs and enterprise teams building agents. | United States | $10M | No |
| Human Archive | Licenses multimodal datasets of real people, recorded on video with synced sensors and then labeled, to robotics developers. | United States | $8.7M | n/a |
| Gradient Health | Medical data company whose Atlas product offers de-identified scans and clinical records to healthcare AI developers. | United States | $8.37M | n/a |
| Aptura | Produces expert training and eval data, benchmarks and RL environments for finance, healthcare and legal AI. | United Kingdom | $6.7M | n/a |
| Cortex AI | Provides workplace video, robot trajectories and supervised robot rollouts that show recovery from failure, for embodied AI labs. | United States | $6.5M | n/a |
| VisionLab AI | Sells real factory video with SOP-linked labels to robotics teams training physical AI. | United States | $6M | n/a |
| Huzzle Labs | Sells RL environments, expert data, evals and custom models for agents that handle real business software. | United Kingdom | $5.98M | n/a |
| Phinity | Builds chip design RL environments and training infrastructure while developing its own chip design agents. | United States | $5.5M | n/a |
| Nyne AI | Nyne.ai, which titles its site 'The Data Factory for AI,' with a broker and referral offer. Payout or referral figures on the site are vendor claims. | United States | $5.3M | Yes |
| Datoric | Licenses robotics data such as first-person video and teleoperation recordings, and runs custom collections to spec. | United States | $4.5M | n/a |
| Troveo AI | Licenses and packages rights-cleared video, audio, text, workflow, gaming and robotics data for AI developers. | United States | $4.5M | No |
| Vetto AI | Sells labs expert-written post-training data, human evaluation work, custom RL environments and verifier-scored benchmarks. | United States | $3.64M | n/a |
| Good Start Labs | RL environments built on video games, plus gameplay training data and public benchmarks. | United States | $3.6M | n/a |
| Paper Instruments | Research shop that fine-tunes language models for knowledge-heavy work and builds the evals, tooling and agent software around them. | United States | $3.6M | n/a |
| Besimple AI | Supplies licensed conversational audio, expert labeling and benchmarks to teams building speech and voice models. | United States | $3.5M | n/a |
| Vibrant Labs | RL-ready simulated environments, tasks and benchmarks for browser, computer and enterprise agents. | United States | $2.5M | n/a |
| Emulated | Provides post-training data, production-style engineering environments and benchmarks for coding and ops agents. | United States | $850K | n/a |
| Abundant | Builds reinforcement learning environments, benchmarks and datasets for labs and companies working on AI agents. | United States | $500K | n/a |
| Aviro | Sells RL environments, post-training data and evals for long-running agents across coding, computer use and office work. | United States | $500K | n/a |
| Coarena | Hosts a public site where people compare two computer-use agents on the same task, and sells the vote data to AI labs. | per its site (San Francisco per its Y Combinator profile) | $500K | n/a |
| Enact | Finds where robot models fail on real hardware, records recovery demos, and fine-tunes and evaluates the models. | United States | $500K | n/a |
| Hillclimb | Frontier lab supplier of math RL environments and fine-tuning examples written by expert problem solvers. | United States | $500K | n/a |
| Magma | Collects approved agent interaction traces and corrections from companies, anonymizes them and licenses them as data. | United States | $500K | n/a |
| Maingen | Builds industrial operations simulations and RL environments, such as running solar portfolios or factories. | United States | $500K | n/a |
| Markov | Sells computer-use datasets and environments built from expert CAD, office, coding and gaming workflows. | United States | $500K | n/a |
| Ooak Data | Y Combinator company that describes turning internal company data into a training ground for AI agents. | France | $500K | No |
| Petrarch | Sources company records under contract, de-identifies them inside the supplier's systems with its own software, and licenses them. | United States | $500K+ | n/a |
| Refresh | Simulated software environments, verifiable evals and real-workflow coding data for agent training. | United States | $500K | n/a |
| TensorTest | Private evals, RL environments and expert data for creative agents, starting with video editing (TimelineBench). | United States | $500K | n/a |
| Traverse | RL environments and data for training models on judgment-heavy, long-horizon professional work. | United States | $500K | n/a |
| Anthromind | Runs managed projects that deliver curated preference and training data, custom model evals and fine-tuning support, using experts. | United States | $200,000 | n/a |
| 3DCH Media | Documentary producer and distributor that licenses its own and partner video archives, raw and edited, to AI companies for training. | per its site | n/a | n/a |
| 6thSense | Records touch, video and motion data from people doing hand tasks with its own wearable sensor and camera, and sells it to robotics teams. | United States | n/a | n/a |
| AI Astrolabe | Data and evaluation vendor centered on Arabic dialects and other underserved languages, covering text, speech, images and safety data. | United States | n/a | n/a |
| Apex Data Sciences | AI unit of Apex CoVantage, started in 2024, selling expert RLHF and SFT data, benchmarks and multimodal annotation. | United States | n/a | n/a |
| Appen | Human data vendor. Labs and product teams buy its labeling, expert review, model evaluation and RL environment work. | Australia | Public | No |
| Archangel Health | Medical AI data vendor offering EHR sandbox RL environments, physician-written answers and rubrics, de-identified clinical data and labeling. | per its site (Y Combinator lists San Francisco) | n/a | n/a |
| Asimov | Films everyday chores in real homes and offices from a first-person view, adds sensor data and labels, and sells it to humanoid robot makers. | United States | n/a | n/a |
| Cint | Survey and respondent network that collects new text, audio and video responses for AI training and validation. | Sweden | n/a | n/a |
| CNTXT AI | Abu Dhabi data shop sourcing, labeling and evaluating training data, including Arabic speech and ranked model responses. | United Arab Emirates | n/a | n/a |
| CollectData | Runs paid collection studies where contributors record video, depth, audio and motion of real tasks using phones and smart glasses. | per its site | n/a | n/a |
| Consequence Labs | Produces custom human data such as workflow demos, speech, expert judgments and labels, recruiting and managing contributors. | n/a | n/a | n/a |
| Curiosity Stream | Nasdaq-listed streaming company (CURI) that also licenses video, audio and code datasets to AI developers. | United States | Public | n/a |
| DARE Labs | Sells Cerebro, software that turns company operational data into training sets, RL data and outcome-graded environments. | United States | n/a | n/a |
| DataOcean AI | Sells stock datasets and custom collection, labeling and evaluation for speech, vision, LLM and driving models. | United States | n/a | n/a |
| DecisionFacts | Runs custom robot data collection with trained operators recording teleoperated and handheld manipulation demos. | per its site | n/a | n/a |
| DeepLLMData | Xlairs Inc. brand making expert training and eval data, including repository-level coding trajectories with tests. | per its site; it also lists Mumbai and Delhi | n/a | n/a |
| Dissei | Finance-focused vendor selling RL environments, benchmarks, transaction data and analyst trajectories. | n/a | n/a | n/a |
| Eigent AI | Open-source multi-agent desktop software maker that also builds custom RL environments and benchmarks for enterprises. | United Kingdom | n/a | n/a |
| Exabite | Supplies large pretraining and post-training corpora, including code, documents and video, plus STEM RL environments. | United States | Self-funded | n/a |
| exponential | Sells multi-agent security environments, security training data and adaptive red team testing. | Germany | n/a | n/a |
| Flevy | Business template marketplace whose data services arm catalogs firms' documents and pitches them to AI labs for licensing. | United States | n/a | Yes |
| Foundry AI | Creates expert-reviewed coding and technical workflow data, RL task suites and private evals. | India | n/a | n/a |
| FS Studio | Flakasoft Inc. brand running robot data collection, including teleoperation floors and handheld gripper demos. | per its site | n/a | n/a |
| G2i | Runs managed coding data projects and places contract engineers inside lab teams for post-training work. | United States | Self-funded | n/a |
| GroundTruth | Pune company licensing benchmarks, including the BFSI-Bench Indian finance regulation test, and running custom evals. | India | n/a | n/a |
| Humyn Labs | Collects and annotates first-person video, motion signals and speech for AI labs and robot makers. | United States | n/a | n/a |
| Innodata | Data engineering firm supplying training data, human feedback, RL environments, evals and red teaming. | United States | n/a | n/a |
| LightWrk | Produces expert reasoning labels, preference signals and eval workflows for robotics and world model teams. | n/a | n/a | n/a |
| Logitly | Hong Kong firm that generates agent training traces with AI and checks them automatically, including DPO trace pairs. | per its site | n/a | n/a |
| LXT | Custom and stock training data, annotation, evaluation and RLHF across text, speech, images and video. | Canada | n/a | n/a |
| Maxinsights | Records and labels first-person video and body motion of people at work for physical AI and robotics customers. | United States | n/a | n/a |
| Metaphi | Builds simulated workplaces, custom or off the shelf, along with datasets and evals for agent training. | United States | n/a | n/a |
| Miro Advisory | Advisory firm that helps established businesses and software companies license their operating records to AI buyers. | United States | n/a | Yes |
| Miso Motion | Runs custom multi-camera video collection of consenting professionals at work, delivered redacted, for robotics teams. | parent company) | n/a | n/a |
| Navaday | Manages specialists who write examples, fix data and compare model answers across professional fields. | n/a | n/a | n/a |
| Olam Labs | Builds multi-agent simulations, behavioral evals and game-based data for studying social and safety behavior. | United States | n/a | n/a |
| OpenTrain AI | Marketplace connecting AI teams with vetted experts for RLHF, evals, red teaming and labeling. | United States | n/a | n/a |
| Originator | Early-stage company with a waitlist, aiming to build data that teaches AI to run legacy software. | n/a | n/a | n/a |
| Parsewave | Writes custom expert datasets, solution traces and evals for coding and professional domain models. | United States | n/a | n/a |
| Pixelstwo | Uses its own touch-sensing hardware to capture dexterous manipulation data, which it sells to robotics teams. | n/a | n/a | n/a |
| pre.dev | Sells graded coding environments and data built from licensed private code repositories. | United States | n/a | n/a |
| Provision | Real-world multimodal data, physical-world evals and deployment support for robotics teams. | n/a | n/a | n/a |
| Proximal | Builds coding tasks, data and RL environments for training software engineering agents. | United States | n/a | n/a |
| Raycaster | Eval programs, data and reproducible environments for agents working in documents and spreadsheets. | United States | n/a | n/a |
| ReasonCore | Expert-curated datasets, RL environments and adaptive evals for labs and enterprises. | n/a | n/a | n/a |
| Redpine | Swedish company licensing non-public data from publishers and research firms and reselling access to AI builders. | per its site | n/a | n/a |
| Rise Data Labs | Human and synthetic data production, labeling and evaluation services for generative AI teams. | United States | n/a | n/a |
| San Francisco Data | Licenses training data built from the logs of companies' AI systems in production. Coding agents come first. | United States | n/a | n/a |
| Sherpa Labs | San Francisco builder of coding and finance RL environments, with tasks written by senior engineers and private equity professionals. | United States | n/a | n/a |
| Sieve | Supplies curated or freshly collected video, audio, images and interaction data, with labels, eval sets and environments. | United States | n/a | n/a |
| Surge | Human-data and RLHF vendor. | United States | n/a | n/a |
| Symbal | Multimodal post-training data, private evals and human-in-the-loop data for labs, plus enterprise agents. | United States | n/a | n/a |
| Tacit Labs | Expert-written life science evals, agent environments and datasets for biology and chemistry work. | United States | n/a | n/a |
| Telegraph Lab | Sources business records for AI training, reviews rights, de-identifies and structures them, and arranges delivery to labs. | United States | n/a | Yes |
| TELUS Digital | TELUS unit supplying expert data, labeling, RL environments, red teaming, evals and physical AI datasets. | Canada | Public | n/a |
| The General Data Company | Brokers anonymized operating records and workflow data from European companies to AI labs. | USA | n/a | Yes |
| Tiki AI | Custom post-training data for frontier teams: coding RL environments, computer-use fine-tuning sets and tool-use environments. | per its site | n/a | n/a |
| Uber AI Solutions | Uber's AI data arm: collection, labeling, specialist RLHF, red teaming and evals for labs and robotics firms. | United States | n/a | n/a |
| Ulam | Math and STEM focused vendor of reasoning data, expert-checked solution traces, benchmarks and RL environments with verifiers. | Poland | n/a | n/a |
| Verita AI | Supplies vetted domain experts and project operations for labeling, feedback, eval and RL environment work. | United States | n/a | n/a |
| Watt | Licenses private operating data for AI training and evaluation, starting with energy and industrial companies. | United States | n/a | n/a |