Domain on the map

Math benchmarks for LLMs and RL environments for math

Mathematical problem solving, from grade-school word problems to competition and research-level questions, usually graded against an exact final answer.

RL environments
5
AI benchmarks
96
Shares entries with
13 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.

Math, 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 math?
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 math, that means mathematical problem solving, from grade-school word problems to competition and research-level questions, usually graded against an exact final answer.
Who builds RL environments for math?
Prime Intellect, Surge AI, Sepal AI, Hillclimb and Ulam.
What are the main math benchmarks for LLMs?
AIME 2025, AIME, APEX, HELM GSM8K, MATH and IMO 2026, among 86 confirmed below.

RL environments 5

  1. Prime Intellect
  2. Surge AI
  3. Sepal AI
  4. Hillclimb
  5. Ulam

AI benchmarks 96

  1. AIME 2025
  2. AIME
  3. APEXMercor
  4. HELM GSM8K
  5. MATHDan Hendrycks et al
  6. IMO 2026
  7. RiemannBench
  8. MATH-500OpenAI
  9. Apex Shortlist
  10. USAMO 2026
  11. ProofBenchVals AI
  12. HMMT 2025MathArena
  13. MathArena ApexETH Zurich SRI Lab
  14. ArxivMathMathArena
  15. IMO 2025
  16. MATH Level 5
  17. GSM8K
  18. AIME 2026MathArena
  19. FrontierMath 2025-02-28 Private
  20. FrontierMath Tier 4 2025-07-01 Private
  21. OTIS Mock AIME 2024-2025
  22. Math RL Tasks
  23. IMOAnswerBenchGoogle DeepMind
  24. FormalRewardBench
  25. IneqMathStanford
  26. Omni-MATHAlibaba, Peking University
  27. PutnamBench
  28. GeoSym-Bench
  29. GSM-8K
  30. HMMT February 2026MathArena
  31. IMO-AnswerBenchGoogle DeepMind
  32. TheoremQA
  33. AIME 2024
  34. FrontierMathEpoch AI
  35. FrontierMath Tier 4Epoch AI
  36. Measuring Autonomous AI Research
  37. General Agent synthetic environment
  38. AutoRO-Bench
  39. CAM-Bench
  40. CNMO 2024
  41. FormInv
  42. FrontierOR
  43. InteractiveBench Math
  44. JEEBench
  45. LinAlg-Bench
  46. LiveMathematicianBench
  47. M3Kang
  48. MathAtlas
  49. Not All Proofs Are Equal
  50. OlympiadBenchOpenBMB
  51. PhiBench
  52. ProofNet
  53. Soohak
  54. SVAMPSatwik Bhattamishra, Navin Goyal, Arkil Patel, NAACL 2021 paper
  55. USAMO25
  56. AIME 2023MathArena
  57. FrontierMath v2Epoch AI
  58. AA AIME 2025Artificial Analysis
  59. Beyond AIME
  60. BRUMO 2025MathArena
  61. AIME26
  62. ArXivMath Jun. 2026MathArena
  63. ArXivMath Aug. 2026
  64. CMath
  65. AIME25
  66. HMMT Feb 2023MathArena
  67. HMMT Feb 2024
  68. HMMT Feb 2025
  69. IPhO 2025
  70. HMMT Nov 2025MathArena
  71. HMMT Feb 2026
  72. MMAnswerBench
  73. AMC_2022_23
  74. Evaluating the Architectural Reasoning Capabilities of LLM Provers via the Obfuscated Natural Number Game
  75. FATE-H
  76. FATE-X
  77. HiddenMath
  78. LeanDojo BenchmarkCalifornia Institute of Technology
  79. MiniF2FOpenAI
  80. Equation-suffix prediction benchmarkDaniel Ranard
  81. FunctionalMATH
  82. MAWPSUniversity of Washington, Koncel-Kedziorski, Roy, Amini
  83. Multilingual MGSM
  84. NL4COP
  85. PolyMath-en
  86. Theory-Scale Auto-Formalization of Logics for Computer Science
10 not yet confirmedListed in the catalog; no line from its own site or paper confirms it yet.
  1. INTELLECT-3 Technical Report
  2. INTELLECT-2
  3. Systematic Reward Hacking study
  4. True Agents Model the World / ECHO
  5. Scaling Agentic RL: 365k+ SWE/terminal/search tasks
  6. SheetBench-50 (released 2025-10-01)
  7. ResearchMath-14K
  8. Universal offline sandbox escape disclosure
  9. SEAL Math
  10. AMC 12

Data 7

  1. Ulambased here
  2. Abundantalso covers it
  3. Bespoke Labsalso covers it
  4. Hillclimbalso covers it
  5. ReasonCorealso covers it
  6. Surge AIalso covers it
  7. Turingalso covers it

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