Every company in the list

Code · Math · Science · Reasoning

An open superintelligence stack: distributed compute, the PRIME-RL training framework, verifiers, and a public Environments Hub positioned as a Hugging Face for RL environments.

Code · Software Engineering

Builds a small number of high-fidelity software-engineering environments and evals for frontier coding agents, with graders that score performance on real SWE tasks.

Long Horizon · Reasoning · Multi-agent

A research-and-infrastructure startup building environments and reasoning data for long-horizon, multi-agent reliability — paired with open community releases.

Data Curation · Evaluation · Reasoning

Evaluation and data-curation tooling for post-training, known for open datasets and reproducible reasoning recipes used widely across the community.

Computer Use · Browser · Finance

YC-backed builder of resettable, high-fidelity sandboxes — a synthetic Salesforce/CRM, for example — plus proprietary datasets for training and evaluating computer- and browser-use agents.

Security · Evaluation · Safety

Runs adversarial arenas and red-teaming programs that probe models and agents for failures, feeding safety evaluations back to developers.

Computer Use · Tooling · Browser

Tooling that wraps arbitrary software — browsers, games, apps — into dockerized, MCP-exposed RL environments. Less an environment catalog than the infrastructure to build them.

Enterprise · Agents · Long Horizon

Builds high-fidelity simulated experiences so enterprises can train and test agents safely before production, targeting accuracy, consistency, and governance.

Long Horizon · Code · RLHF · Evaluation

Builds long-horizon coding and agent environments, RLHF datasets, and evaluation benchmarks — pairing engineered environments with vetted human-expert feedback. Early-stage, but unusually broad for its size.

Long Horizon · Evaluation · Safety

Builds attention-grabbing real-world agent benchmarks (vending machines, radio stations, robot control) and develops protocols for controlling autonomous AI systems.

Code · Finance

Benchmark-style environments around code and finance workflows, with a focus on practical, real-world task framing rather than toy problems.

Code

Code-execution and code-evaluation environments plus high-quality coding data, built for fast, iterative model-training loops.

Browser · Enterprise

Blends browser-interaction environments with enterprise workflow simulation for agent training and evaluation.

Finance · Legal · Evaluation

Builds rigorous, domain-specific benchmarks for high-stakes professional work — law, finance, tax — where generic evals miss the nuance.

Finance · Enterprise · Evaluation

Designs expert-grade evaluation environments for complex professional tasks, partnering on public benchmarks like SheetBench.

Browser

A browser-native environment builder focused on web navigation and interaction tasks for RL and evaluation.

Human Data · RLHF · Multi-domain

A bootstrapped human-data powerhouse working with every major lab, now offering RL environments alongside its core data business.

Data Labeling · Agents · Multi-domain

The data-labeling giant of the chatbot era, now retooling for agents and environments through its Forge offering.

Code · Healthcare · Law · Multi-domain

An expert-marketplace and human-data company now pitching RL environments for domain-specific work — coding, healthcare, law.