ProofBench
Full written proofs graded for rigour (Vals AI).
- Published by
- Vals AIdata via Epoch AI Benchmarking Hub
- Category
- Maths
- Index weight
- 0.5
- Models
- 61
- Data as of
- 9 Sept 2026
CC BY 4.0 — Epoch AI, ‘AI Benchmarking Hub’, epoch.ai/benchmarks. Mirrored boards credit their original publishers.
- 1Claude Fable 5.1100.0%
- 2GPT-6 Astra99.0%
- 3Claude Opus 599.0%
- 4Claude Fable 595.0%
- 5Kimi K387.0%
- 6GPT-5.6 Sol83.0%
- 7Claude Sonnet 577.0%
- 8GPT-5.6 Terra74.0%
- 9Claude Opus 4.869.0%
- 10GPT-5.6 Luna60.0%
- 11Gemini 3.7 Flash58.0%
- 12Qwen3 858.0%
- 13GPT-5.456.0%
- 14DeepSeek V4 Flash56.0%
- 15Claude Opus 4.754.0%
61 of 61
| # | |||||
|---|---|---|---|---|---|
| 1 |  Claude Fable 5.1Anthropic | 100.0% | – | 70.2 | – |
| 2 |  GPT-6 AstraOpenAI | 99.0% | – | 69.9 | – |
| 3 | Claude Opus 5Anthropic | 99.0% | – | 66.8 | – |
| 4 | Claude Fable 5Anthropic | 95.0% | – | 65.8 | – |
| 5 |  Kimi K3Moonshot AIopen | 87.0% | kimi/kimi-k3 | 66.5 | – |
| 6 | GPT-5.6 SolOpenAI | 83.0% | – | 65.5 | – |
| 7 | Claude Sonnet 5Anthropic | 77.0% | – | 50.7 | – |
| 8 | GPT-5.6 TerraOpenAI | 74.0% | – | 64.2 | – |
| 9 | Claude Opus 4.8Anthropic | 69.0% | – | 59.4 | – |
| 10 | GPT-5.6 LunaOpenAI | 60.0% | – | 56.5 | – |
| 11 |  Gemini 3.7 FlashGoogle | 58.0% | – | 61.7 | – |
| 12 |  Qwen3 8Alibaba | 58.0% | – | – | – |
| 13 | GPT-5.4OpenAI | 56.0% | – | 61.3 | – |
| 14 |  DeepSeek V4 FlashDeepSeekopen | 56.0% | 0731 | 59.1 | – |
| 15 | Claude Opus 4.7Anthropic | 54.0% | – | 59.3 | – |
| 16 |  Grok 4.6xAI | 51.0% | grok/grok-4.6 | 63.6 | – |
| 17 | GPT-5.5OpenAI | 50.0% | – | 63.7 | – |
| 18 | DeepSeek V4 ProDeepSeekopen | 50.0% | 0813 | 59.7 | – |
| 19 | Claude Opus 4.6Anthropic | 50.0% | – | 54.1 | – |
| 20 |  GLM-5.3Zhipu AIopen | 49.0% | – | 59.9 | – |
| 21 | Gemini 3.8 FlashGoogle | 48.0% | – | 62.4 | – |
| 22 | Claude Sonnet 4.6Anthropic | 45.0% | – | 49.7 | – |
| 23 |  Muse Spark 1.2Meta | 43.0% | meta/muse_spark_1_2 | 62.6 | – |
| 24 | Muse Spark 1.1Meta | 39.0% | meta/muse_spark_1_1 | 64.9 | – |
| 25 | Gemini 3.6 FlashGoogle | 36.0% | – | 60.7 | – |
| 26 | Claude Opus 4.5Anthropic | 36.0% | – | 58.0 | – |
| 27 | GLM-5.2Zhipu AIopen | 35.0% | – | 57.6 | – |
| 28 | Gemini 3.5 FlashGoogle | 31.0% | – | 60.4 | – |
| 29 | Grok 4.5xAI | 31.0% | – | 54.2 | – |
| 30 | Gemini 3.1 ProGoogle | 26.0% | – | 63.9 | – |
| 31 | Qwen3 7Alibaba | 26.0% | – | 58.9 | – |
| 32 | GLM-5.1Zhipu AIopen | 22.2% | zai/glm-5.1 | 60.1 | – |
| 33 | MiMo-V2.5-ProXiaomiopen | 22.0% | – | 59.0 | – |
| 34 | GLM-5.3-FlashZhipu AIopen | 21.0% | – | 55.6 | – |
| 35 | GPT-5.4 miniOpenAI | 21.0% | – | 46.0 | – |
| 36 | Gemini 3 ProGoogle | 20.0% | – | 61.6 | – |
| 37 | Claude Sonnet 4.5Anthropic | 19.0% | – | 52.3 | – |
| 38 |  MiniMax-M3MiniMaxopen | 18.0% | minimax/MiniMax-M3 | 56.9 | – |
| 39 | GPT-5OpenAI | 18.0% | – | 54.5 | – |
| 40 | Muse SparkMeta | 17.0% | – | 65.4 | – |
| 41 | Kimi K2.6Moonshot AIopen | 16.0% | kimi/kimi-k2.6 | 60.6 | – |
| 42 | Qwen 3.8 27BAlibabaopen | 16.0% | – | 57.1 | – |
| 43 | DeepSeek V4 ProDeepSeekopen | 16.0% | – | 56.1 | – |
| 44 | GPT-5.2OpenAI | 15.0% | – | 57.3 | – |
| 45 | Gemini 3 FlashGoogle | 15.0% | – | 56.6 | – |
| 46 | Grok 4.20xAI | 14.0% | – | 54.3 | – |
| 47 | Gemini 3.5 Flash LiteGoogle | 13.0% | – | 53.5 | – |
| 48 | GPT-5 nanoOpenAI | 12.0% | – | 45.9 | – |
| 49 | Grok 4.3xAI | 11.0% | – | 50.1 | – |
| 50 | GPT-5 miniOpenAI | 9.0% | – | 52.3 | – |
| 51 |  Mistral Medium (latest)Mistral AI | 9.0% | medium, 2604 | 45.8 | – |
| 52 | GPT-5.1-Codex-MaxOpenAI | 9.0% | – | – | – |
| 53 | Inkling-SmallThinking Machinesopen | 6.0% | – | 55.0 | – |
| 54 | GLM-4.7Zhipu AIopen | 6.0% | zai/glm-4.7 | 54.9 | – |
| 55 | GPT-5.4 nanoOpenAI | 5.0% | – | 48.3 | – |
| 56 | Grok 4.1xAI | 4.0% | – | 55.9 | – |
| 57 | MiniMax-M2.5MiniMaxopen | 4.0% | minimax/MiniMax-M2.5 | 53.5 | – |
| 58 | MiniMax-M2.7MiniMaxopen | 3.0% | minimax/MiniMax-M2.7 | 56.5 | – |
| 59 |  Nemotron 3 UltraNVIDIA | 2.0% | – | 48.2 | – |
| 60 |  Laguna M.1Poolsideopen | 0.0% | poolside/laguna-m.1 | 37.5 | – |