RL List

The 2026 List

A ranked list of the companies building the environments AI learns in.

Reinforcement learning environments are where frontier models are trained, stressed, and measured. RL List catalogues the labs and companies shipping them — scored on a single public rubric.

16
Companies indexed
05
Open-source projects
03
Scoring dimensions

01 · Leaderboard

The current ranking

Full profiles →
  1. 01Prime Intellectthe open-source backbone of the category8.80
  2. 02Mechanizedeep, narrow, elite7.58
  3. 03General Reasoningresearch-grade, open-leaning7.15
  4. 04Bespoke Labsopen data and reproducible recipes7.15
  5. 05Halluminatecomputer-use, done deterministically7.03
  6. 06Gray Swan AIadversarial evaluation arenas6.85
  7. 07HUDthe wrapper layer6.65
  8. 08Veris AIthe enterprise environment layer6.53
  9. 09Huzzle Labsenvironments + human data + evals in one stack6.33
  10. 10Andon Labsagent benchmarks with a sense of humor6.15
  11. 11AfterQuerybenchmarks with real-task framing6.10
  12. 12Datacurvecode execution, tight loops6.10
  13. 13Platobrowser meets enterprise workflow6.10
  14. 14Vals AIdomain benchmarks for regulated work6.00
  15. 15Sepal AIexpert evaluation environments6.00
  16. 16Matricesbrowser-native5.68

Outside the ranking

Also building RL environments

These companies built their reputations in human data and labeling and now offer RL environments too — but environments aren't their core focus, so they sit outside the ranking.

Adjacent · Infrastructure

Adjacent: execution infrastructure

These companies build the sandbox and runtime layer that RL environments execute on — fast, isolated, resettable compute. They're not environment vendors themselves, but they're the substrate the category runs on.

02 · Method

How we score

Every ranked company is graded against the same three dimensions, weighted into a single 0–10 score. The rubric is public and applies to all listings — no exceptions.

Read the full method →

Growth & momentum

50%

Funding, hiring, shipping cadence, and how the trajectory has moved over the last few quarters.

Traction & credibility

35%

Named frontier-lab customers, founder pedigree, research output, and technical depth.

Accessibility

15%

Open-source releases, public docs, self-serve access vs. closed enterprise.

03 · Writing

Latest guides

All guides →
2026-05-22

How to Choose an RL Environment Vendor

A buyer's framework for picking an RL environment vendor — what to ask, what to ignore, and how to match a vendor's strengths to the kind of agent you're actually trying to train.

Read →2026-05-22

RL Environments for Coding Agents

Coding is the most commercially active RL environment domain. Here's how code environments work, what makes a good one, and the companies building them in 2026.

Read →2026-05-22

RL Environments for Browser & Computer-Use Agents

Computer-use is where agents finally hit the real world: clicking buttons, filling forms, navigating apps. Here's how the environments work and who builds them.

Read →