Domain on the map

RL environments and benchmarks for computer use agents

Operating real software through its interface: clicking, typing and navigating desktops, browsers and apps the way a person at a keyboard would.

RL environments
21
AI benchmarks
72
Shares entries with
16 categories

Snapshot 2026-11-12. How categories are built · Where the data comes from

The galaxy could not load. See every vendor on the board.

Computer use, live from the universe. Hover a star for its name; click to open it. Open it in the universe

In short

What are RL environments for computer use agents?
A simulated workplace where an AI agent practises a task (a code repository, a browser, business software) and an automatic grader scores each attempt. Labs train on thousands of these attempts. Bought to improve models. For computer use agents, that means operating real software through its interface: clicking, typing and navigating desktops, browsers and apps the way a person at a keyboard would.
Who builds RL environments for computer use agents?
Scale AI, Snorkel AI, AfterQuery, Surge AI, Fleet AI and Deeptune, among 19 confirmed below.
What are the main computer use benchmarks?
OSWorld2.0, Vending-Bench, BlenderBench, Web Bench (2025-05-30; ~5, SWE-Bench Pro v1 (2025-09-19) and v2 and MAGI, among 37 confirmed below.

RL environments 21

  1. Scale AI
  2. Snorkel AI
  3. AfterQuery
  4. Surge AI
  5. Fleet AI
  6. Deeptune
  7. Halluminate
  8. Andon Labs
  9. Collinear
  10. Chakra Labs
  11. Huzzle Labs
  12. HUD
  13. Plato
  14. Refresh
  15. BenchFlow
  16. Cua
  17. Markov
  18. Vetto AI
  19. Habitat Inc
2 not yet confirmedListed in the catalog; no line from its own site or paper confirms it yet.
  1. Originator
  2. Matrices

AI benchmarks 72

  1. OSWorld2.0XLANG Lab, University of Hong Kong
  2. Vending-Bench
  3. BlenderBench
  4. Web Bench (2025-05-30; ~5
  5. SWE-Bench Pro v1 (2025-09-19) and v2
  6. MAGIChakra Labs
  7. DrugDiscoveryBench with Phylo
  8. READY or Not enterprise agent deployment benchmarksScale AI
  9. RSI Bench
  10. PRBench
  11. Westworld v1 simulated-web environments (2025-11-23):
  12. BrowserBench (2025-11-22):
  13. Legora BARAfterQuery
  14. TraitBasis / Impatient Users Confuse AI AgentsCollinear
  15. Vending-Bench 2 / Vending-Bench ArenaAndon Labs
  16. Blueprint-Bench 2, 2026-05Andon Labs
  17. Drone-Bench, with Anthropic Project Pilot, 2026-07Andon Labs
  18. ClawsBenchBenchFlow
  19. Robo UseBenchFlow
  20. FrontierPhysics v0.1BenchFlow
  21. Benchmarking Agents in Insurance Underwriting Environments
  22. BigLaw Bench: ResearchHarvey AI
  23. Westworld Finance Diligence Bench (2026-08-11):
  24. DealTrace Bench (2026-05/06):
  25. VADER
  26. Blueprint-Bench
  27. Butter-Bench
  28. LibraryDesignBench
  29. SlopCodeBench
  30. Assemble Bench (robotics
  31. SheetBench-50 (with Sepal AI) - 2025-10-01
  32. Cua-Bench EDA Leaderboard
  33. Gemini 3.5 Flash computer-use evaluation on Cua-BenchCua
  34. InsureBench: insurance underwriting and claims benchmark
  35. SWE Atlas: Codebase QnA, Test Writing, Refactoring
  36. Andon FM
  37. TUA-Bench
35 not yet confirmedListed in the catalog; no line from its own site or paper confirms it yet.
  1. CUA data
  2. OSWorld Windows tasks on Daytona Windows sandboxes
  3. Agent-environment papers
  4. The Valley of Code Reasoning
  5. Cats Confuse Reasoning LLM
  6. Self-rationalization improves LLM as a fine-grained judge
  7. AI bosses studies, parts 1 and 2
  8. env0 SFT study on SkillsBench/ClawsBench
  9. GenesisBench
  10. FinQA / SnorkelFinance environment and benchmark
  11. RIFT / Hitchhiker's Guide to Rubric Quality
  12. Shrinking the Generation-Verification Gap with Weak Verifiers
  13. Snorkel: Rapid Training Data Creation with Weak Supervision
  14. Simulated-to-Real Transfer for Web Agents (2026-07-27):
  15. Project Vend
  16. Cheating in Drone-Bench
  17. LiveWeb samples (with Browserbase) - 2026-09-22/28
  18. Production-debugging RL environment (hud-ops-diagnostics) - 2026-01-20
  19. Dojo-Bench-Mini public task set on Hugging Face
  20. Pango crowdsourced computer-use trajectory datasets on Hugging Face
  21. GLADOS-1
  22. Gauntlet 4K RLVR dataset, access by request
  23. OpenEnv Visual Memory and Spreadsheet environments on Hugging Face
  24. Printing Machines (2026-01-28)
  25. No peer-reviewed paper authored by Fleet found. Harbor
  26. Editorial and infrastructure posts on chakra.dev/blog
  27. HuzzleWorld computer-use models
  28. Searching Over Sandbox States for Coding Agents
  29. CaptchaMind
  30. Scale Labs (launched 2026-03-09
  31. PostTrain Arena
  32. Learning from Less: RLVR in Low Data Regimes
  33. computer-1 computer-use harness in Harbor
  34. Rubric-based RL papers
  35. Kubernetes Is the Wrong Primitive for Diverse Workloads

Data 17

  1. Markovbased here
  2. Tolokabased here
  3. AfterQueryalso covers it
  4. Aviroalso covers it
  5. Chakra Labsalso covers it
  6. Collinearalso covers it
  7. Contra Labsalso covers it
  8. HUDalso covers it
  9. Huzzle Labsalso covers it
  10. Normalalso covers it
  11. Patronus AIalso covers it
  12. Refreshalso covers it
  13. Rise Data Labsalso covers it
  14. Snorkel AIalso covers it
  15. Surge AIalso covers it
  16. Thetaalso covers it
  17. Trajectory Labsalso covers it

Next: pick another domain. Tags come from the RL Research daily scrape, canonicalized nightly.