Skip to content

FrontierMath v2 Tier 4

Epoch AI's corrected v2 Tier 4 expansion, a separate set of exceptionally difficult research-level mathematics problems evaluated with Python-enabled iterative reasoning.

51 models

Higher is better.

  1. 1GPT-6.1 SolOpenAI100.000
  2. 2GPT-6 AstraOpenAI97.600
  3. 3Claude Opus 5.5Anthropic95.000
  4. 4Claude Fable 5Anthropic90.200
  5. 5GPT-6 SolOpenAI90.000
  6. 6Claude Fable 5.1Anthropic87.805
  7. 7GPT-5.6 SolOpenAI83.000
  8. 8Claude Sonnet 5.5Anthropic80.488
  9. 9Gemini 4 ArgonGoogle78.049
  10. 10GPT-5.5 ProOpenAI78.049
  11. 11Claude Opus 5Anthropic73.171
  12. 12GPT-5.5OpenAI72.500
  13. 13GPT-5.6 TerraOpenAI68.300
  14. 14GPT-5.4 ProOpenAI58.537
  15. 15GPT-5.6 LunaOpenAI58.500
  16. 16Claude Opus 4.8Anthropic56.098
  17. 17GPT-6 LunaOpenAI56.098
  18. 18GPT-5.4OpenAI49.000
  19. 19Qwen3.8 MaxAlibaba46.341
  20. 20Claude Haiku 5.5Anthropic46.341
  21. 21Muse Spark 1.3Meta46.341
  22. 22Gemini 3.7 FlashGoogle36.585
  23. 23Qwen3.7 MaxAlibaba34.146
  24. 24Grok 4.6xAI31.707
  25. 25Claude Opus 4.7Anthropic31.707
  26. 26GPT-5.2OpenAI31.700
  27. 27GLM-5.3Z.AI29.268
  28. 28Claude Sonnet 5Anthropic29.268
  29. 29GLM-5.2Z.AI29.268
  30. 30DeepSeek V4.1 Flash 1017DeepSeek26.829
  31. 31Gemini 3.1 ProGoogle26.829
  32. 32Claude Opus 4.6Anthropic26.829
  33. 33Gemini 3.5 FlashGoogle26.829
  34. 34Kimi K2.6Moonshot AI25.641
  35. 35Grok 4.5xAI24.390
  36. 36Gemini 3.8 FlashGoogle21.951
  37. 37Gemini 3.6 FlashGoogle21.951
  38. 38GLM-5.3-FlashZ.AI17.073
  39. 39Gemini 3 FlashGoogle17.073
  40. 40Inkling-SmallThinking Machines Lab17.073
Higher is better. Showing the top 40 of 51. Bar colour: blue proprietary, orange open weights, green other.

About this benchmark

Metric
Python-enabled iterative mathematical problem solving
Tasks
41 private extreme-difficulty mathematics problems
Authors
Epoch AI (2026)

Scores are as reported by model developers or evaluators and may use different settings (for example reasoning effort or tools). Compare with care.

All results

1GPT-6.1 Sol OpenAI100.000
2GPT-6 Astra OpenAI97.600
3Claude Opus 5.5 Anthropic95.000
4Claude Fable 5 Anthropic90.200
5GPT-6 Sol OpenAI90.000
6Claude Fable 5.1 Anthropic87.805
7GPT-5.6 Sol OpenAI83.000
8Claude Sonnet 5.5 Anthropic80.488
9Gemini 4 Argon Google78.049
10GPT-5.5 Pro OpenAI78.049
11Claude Opus 5 Anthropic73.171
12GPT-5.5 OpenAI72.500
13GPT-5.6 Terra OpenAI68.300
14GPT-5.4 Pro OpenAI58.537
15GPT-5.6 Luna OpenAI58.500
16Claude Opus 4.8 Anthropic56.098
17GPT-6 Luna OpenAI56.098
18GPT-5.4 OpenAI49.000
19Qwen3.8 Max Alibaba46.341
20Claude Haiku 5.5 Anthropic46.341
21Muse Spark 1.3 Meta46.341
22Gemini 3.7 Flash Google36.585
23Qwen3.7 Max Alibaba34.146
24Grok 4.6 xAI31.707
25Claude Opus 4.7 Anthropic31.707
26GPT-5.2 OpenAI31.700
27GLM-5.3 Z.AI29.268
28Claude Sonnet 5 Anthropic29.268
29GLM-5.2 Z.AI29.268
30DeepSeek V4.1 Flash 1017 DeepSeek26.829
31Gemini 3.1 Pro Google26.829
32Claude Opus 4.6 Anthropic26.829
33Gemini 3.5 Flash Google26.829
34Kimi K2.6 Moonshot AI25.641
35Grok 4.5 xAI24.390
36Gemini 3.8 Flash Google21.951
37Gemini 3.6 Flash Google21.951
38GLM-5.3-Flash Z.AI17.073
39Gemini 3 Flash Google17.073
40Inkling-Small Thinking Machines Lab17.073
41Grok 4.7 xAI17.073
42Grok 4.3 xAI14.634
43Kimi K2.7 Code Moonshot AI12.195
44GPT-5.4 nano OpenAI12.195
45GPT-5.4 mini OpenAI9.756
46Claude Opus 4.5 Anthropic4.878
47Inkling Thinking Machines Lab4.878
48o4-mini (high) OpenAI4.878
49Claude Sonnet 4.5 Anthropic2.439
50Gemini 3.5 Flash-Lite Google0.000
51Gemini 2.5 Pro Google0.000