Math: AI training data and RL environment vendors
Vendors with public evidence of math work. This is a shortlist by domain, not a quality ranking; multi-domain vendors appear here only when this specific domain is supported.
Math: AI training data and RL environment vendors
15 vendors. 5 state they have a SOC 2 Type I or Type II report and 4 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 |
|---|---|---|---|---|---|---|
| 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 | |
| ExabiteSupplies large-scale pretraining and post-training datasets, including code and web, document, and video data, plus verifiable STEM RL environments and custom data sourcing for frontier AI labs. | Training data, RL envs | Coding, Science, Math, Enterprise workflows | Open data | — | 2026-10-13 | |
| HillclimbHillclimb creates custom mathematics-focused reinforcement-learning environments and supervised fine-tuning data from expert problem-solving, selling them to frontier AI labs to train models and research agents. | RL envs, Human data, Training data | Math | Expert-made | $500K | 2026-09-29 | |
| InnodataInnodata provides AI developers and enterprises with training data, human feedback, annotation, reinforcement-learning environments, model evaluations, and red-teaming services for building and improving AI models and agents. | RL envs, Human data, Training data, Evals, Annotation | Coding, Computer use, Browser, Enterprise workflows, Finance, Legal, Medical, Science, Math, Security, Robotics and physical, Multi-domain | Customer data, Expert-made, Synthetic | — | 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 | |
| ReasonCoreReasonCore provides expert-curated AI training datasets, reinforcement-learning environments, and adaptive evaluations for frontier AI labs, emerging AI labs, and enterprises developing or deploying advanced models. | RL envs, Training data, Evals, Human data, Annotation | Robotics and physical, Science, Coding, Finance, Math, Enterprise workflows | Customer data, Expert-made | — | 2026-09-29 | |
| SuperAnnotateSuperAnnotate sells AI data software and managed expert services to AI labs and developers, covering multimodal annotation, training data, model and agent evaluations, and reinforcement-learning environments. | RL envs, Human data, Evals, Annotation, Training data | Multi-domain, Medical, Legal, Science, Coding, Browser, Computer use, Robotics and physical, Math, Enterprise workflows, Finance | Expert-made | $67.5M | 2026-09-29 | |
| SurgeSurge sells AI labs and developers expert-built training data, human feedback, evaluation benchmarks, and interactive reinforcement-learning environments to train and assess AI models and agents. | RL envs, Training data, Evals, Human data, Annotation | Multi-domain, Coding, Enterprise workflows, Finance, Medical, Science, Math, Legal | Expert-made | — | 2026-09-29 | |
| SymbalSymbal supplies AI labs with multimodal post-training datasets, private evaluations and benchmarks, and human-in-the-loop data; it also builds enterprise agents using customers’ first-party workflows. | Training data, Evals, Simulation | Multi-domain, Science, Math, Finance, Medical, Legal, Robotics and physical, Coding | Customer data | — | 2026-09-29 | |
| TELUS DigitalAcquired by TELUSTELUS Digital supplies AI labs and developers with expert-built training data, annotations, agentic RL environments, red-team testing and evaluations, plus custom multimodal and physical-AI datasets and managed data-collection services. | RL envs, Human data, Training data, Evals, Annotation | Coding, Enterprise workflows, Finance, Legal, Medical, Math, Robotics and physical, Science, Security, Other, Multi-domain | Expert-made, Customer data | — | 2026-10-13 | |
| TolokaToloka supplies AI labs and developers with human-expert and synthetic training data, agent environments, evaluations, and red-teaming to build and assess AI models and agents. | RL envs, Human data, Evals, Training data, Annotation | Multi-domain, Coding, Computer use, Enterprise workflows, Math, Science, Robotics and physical | Synthetic | $72M | 2026-09-29 | |
| TuringTuring supplies AI labs with training data, evaluations, and RL environments for coding, enterprise knowledge work, and STEM. It also helps large enterprises build and deploy AI agents in important workflows. | RL envs, Training data, Evals, Human data | Multi-domain, Coding, Enterprise workflows, Science, Math, Finance, Legal, Medical, Computer use, Security | Customer data | $225M | 2026-10-13 | |
| UlamUlam sells reasoning datasets, expert-reviewed traces, benchmarks, and verifier-backed RL environments—primarily for mathematics and STEM—to AI labs and developers. | RL envs, Training data, Evals, Human data | Coding, Science, Math, Multi-domain | Synthetic | — | 2026-10-13 | |
| Vals AIVals AI sells evaluation software, expert-built benchmarks, and review tools to AI labs and enterprise engineering teams testing models and AI applications on practical, domain-specific tasks. | Evals, RL infra | Multi-domain, Finance, Legal, Medical, Coding, Math | Expert-made | $45M | 2026-09-29 | |
| VmaxVmax markets Campaign, a self-play task-generation and RL platform that uses developers’ proprietary data and evaluations to create training environments and refine agents on new task examples. | RL envs, RL infra, Evals | Coding, Math | Customer data | $250K | 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.