Games: AI training data and RL environment vendors
Vendors with public evidence of games work. This is a shortlist by domain, not a quality ranking; multi-domain vendors appear here only when this specific domain is supported.
Games: AI training data and RL environment vendors
10 vendors. 2 state they have a SOC 2 Type I or Type II report and 2 state they can run in a customer's cloud or on-prem. These are sourced statements, not independent verification. Pending and unspecified SOC 2 claims are excluded.
— means not publicly listed or not confirmed; it never means zero or “No”. Checked is the date a company’s facts were last checked, and each fact on its profile links to its source.
| Compare | Vendor | Offerings | Domains | Data sources | Funding | Checked |
|---|---|---|---|---|---|---|
| Chakra LabsChakra sells computer-use reinforcement-learning environments, human interaction trajectories, and evaluation resources that AI researchers use to train and test agents. | RL envs, Training data, Evals | Computer use, Enterprise workflows, Games, Creative | App replicas, Expert-made | $10.1M | 2026-10-13 | |
| ClaruClaru provides custom and licensed video, robotics demonstrations, and human annotations to AI labs building robotics, embodied AI, and world models. | Training data, Human data, Annotation, Evals | Robotics and physical, Games, Creative, Multi-domain, Coding, Legal, Medical, Math, Science, Security, Other | Licensed data, Expert-made, Synthetic, Open data | $170M | 2026-10-13 | |
| DatoricDatoric licenses robotics and physical-AI training data (egocentric video, manipulation and teleoperation recordings), runs custom collections to a buyer's specification, and also offers computer-use, voice and gameplay datasets. | Training data, Human data | Computer use, Browser, Enterprise workflows, Robotics and physical, Games, Multi-domain, Other | Expert-made | $4.5M | 2026-10-13 | |
| Good Start LabsGood Start Labs builds game-based reinforcement-learning environments, turns gameplay into human and synthetic training data, and publishes evaluation benchmarks for AI labs. It also helps game publishers capture gameplay data and put AI agents into their games. | RL envs, Human data, Evals, Training data | Games | Customer data, Licensed data | $3.6M | 2026-09-29 | |
| MarkovMarkov sells computer-use datasets and task environments—including expert CAD, productivity, coding, and gaming workflows—to AI labs and developers training or evaluating desktop agents. | RL envs, Training data, Evals, Human data | Multi-domain, Computer use, Coding, Browser, Games | Expert-made | $500K | 2026-09-29 | |
| MidcenturyMidcentury sells real-world human-action and gameplay datasets, along with cloud simulation tools, to robotics developers, AI labs, and enterprises training and evaluating physical AI systems. | Training data, RL envs, Simulation, Human data | Robotics and physical, Games | Licensed data | $15M | 2026-09-29 | |
| Olam LabsOlam Labs builds simulated multi-agent environments, behavioral evaluations and game-derived datasets for AI labs and researchers training or evaluating models’ social, agentic and safety capabilities. | RL envs, Training data, Evals, Simulation | Coding, Computer use, Enterprise workflows, Finance, Games, Security, Multi-domain | — | — | 2026-10-13 | |
| Prime IntellectPrime Intellect sells AI labs and developers an open stack for reinforcement-learning environments, model evaluations, hosted training, inference, sandboxes, and GPU compute to build and improve agentic models. | RL envs, Evals, RL infra | Multi-domain, Math, Coding, Games | — | $150M | 2026-09-29 | |
| QuesmaQuesma publishes coding-agent benchmarks and builds RL training environments for frontier labs; it also markets tools that help development teams understand agent costs, workflows and outcomes. | RL envs, Evals | Coding, Security, Games | — | $2.3M | 2026-10-06 | |
| SieveSieve supplies AI labs and model developers with curated or custom-collected video, audio, image, and interaction data, plus annotations, evaluation sets, and interactive environments. | Training data, Evals, RL envs, Human data | Creative, Computer use, Robotics and physical, Games, Multi-domain | Licensed data, Expert-made | — | 2026-10-13 |
No vendor matches every filter. Filters in different groups must all match. Remove one to widen the list:
— not publicly listed or not confirmed. Open a profile for each fact’s source. The register is unranked; sorting only reorders.
How to read domain and SOC 2 labels
Multi-domain means work across several task areas, not every area. Vendors appear in a specific domain filter only when that area has supporting evidence. SOC 2 entries are vendor statements, not independently verified audits: Unspecified means claimed without a stated type; Pending means in progress; Not publicly listed means we found no public statement. None of those means a completed Type I/II report. SOC 2 and deployment details remain on profiles and in comparison.