ProofBench
Write rigorous mathematical proofs.
Agent Index weight
0%
shown for context only
Models tested
84
from the benchmark’s leaderboard and reports
Results
| # | Model | Score | Normalised | Harness | Cost/task | Date |
|---|---|---|---|---|---|---|
| 1 | Claude Fable 5.1 Supported Anthropic | 100.0% | source | — | — | |
| 2 | Claude Fable 5.2 Supported Anthropic | 100.0% | source | — | — | |
| 3 | Claude Opus 5.5 Supported Anthropic | 100.0% | source | — | — | |
| 4 | Claude Sonnet 5.5 Estimated Anthropic | 100.0% | source | — | — | |
| 5 | Claude Opus 5 Supported Anthropic | 99.0% | source | — | — | |
| 6 | Gemini 4 Argon Supported Google | 99.0% | source | — | — | |
| 7 | GPT-6.1 Sol Estimated OpenAI | 99.0% | source | — | — | |
| 8 | GPT-6 Astra Supported OpenAI | 99.0% | source | — | — | |
| 9 | Claude Fable 5 Supported Anthropic | 95.0% | source | — | — | |
| 10 | Kimi K3.1 Pro Moonshot AI | 91.0% | source | — | — | |
| 11 | Claude Haiku 5.5 Estimated Anthropic | 89.0% | source | — | — | |
| 12 | Kimi K3 Supported Moonshot AI | 87.0% | source | — | — | |
| 13 | Muse Wonder Estimated Meta | 84.0% | source | — | — | |
| 14 | GPT-5.6 Sol Supported OpenAI | 83.0% | source | — | — | |
| 15 | GPT-6 Sol Estimated OpenAI | 83.0% | source | — | — | |
| 16 | Claude Sonnet 5 Supported Anthropic | 77.0% | source | — | — | |
| 17 | Tencent Hy4 preview Tencent | 75.0% | source | — | — | |
| 18 | GPT-5.6 Terra Estimated OpenAI | 74.0% | source | — | — | |
| 19 | MiMo-V2.6-Pro Xiaomi | 70.0% | source | — | — | |
| 20 | Claude Opus 4.8 Supported Anthropic | 69.0% | source | — | — | |
| 21 | GPT-6 Luna OpenAI | 64.0% | source | — | — | |
| 22 | MiMo-V2.6-Flash Xiaomi | 63.0% | source | — | — | |
| 23 | GPT-5.6 Luna Estimated OpenAI | 60.0% | source | — | — | |
| 24 | Gemini 3.7 Flash Estimated Google | 58.0% | source | — | — | |
| 25 | Muse Spark 1.3 Estimated Meta | 58.0% | source | — | — | |
| 26 | Qwen3.8 Max Preview Estimated Alibaba (Qwen) | 58.0% | source | — | — | |
| 27 | DeepSeek V4 Flash 0731 DeepSeek | 56.0% | source | — | — | |
| 28 | GPT-5.4 Supported OpenAI | 56.0% | source | — | — | |
| 29 | Claude Opus 4.7 Supported Anthropic | 54.0% | source | — | — | |
| 30 | DeepSeek V4.1 Flash Estimated DeepSeek | 54.0% | source | — | — | |
| 31 | Grok 4.6 Supported xAI | 51.0% | source | — | — | |
| 32 | Claude Opus 4.6 Supported Anthropic | 50.0% | source | — | — | |
| 33 | DeepSeek V4 Pro 0813 DeepSeek | 50.0% | source | — | — | |
| 34 | GPT-5.5 Supported OpenAI | 50.0% | source | — | — | |
| 35 | GLM-5.3 Supported Z.ai (Zhipu) | 49.0% | source | — | — | |
| 36 | Gemini 3.8 Flash Supported Google | 48.0% | source | — | — | |
| 37 | Claude Sonnet 4.6 Supported Anthropic | 45.0% | source | — | — | |
| 38 | Muse Spark 1.2 Estimated Meta | 43.0% | source | — | — | |
| 39 | Step 5 Preview stepfun | 42.0% | source | — | — | |
| 40 | Muse Spark 1.1 Estimated Meta | 39.0% | source | — | — | |
| 41 | Claude Opus 4.5 (latest) Supported Anthropic | 36.0% | source | — | — | |
| 42 | Gemini 3.6 Flash Estimated Google | 36.0% | source | — | — | |
| 43 | GLM-5.2 Supported Z.ai (Zhipu) | 35.0% | source | — | — | |
| 44 | Gemini 3.5 Flash Supported Google | 31.0% | source | — | — | |
| 45 | Grok 4.5 Estimated xAI | 31.0% | source | — | — | |
| 46 | Gemini 3.1 Pro Preview Supported Google | 26.0% | source | — | — | |
| 47 | Grok 4.7 Estimated xAI | 26.0% | source | — | — | |
| 48 | Qwen3.7 Max Alibaba (Qwen) | 26.0% | source | — | — | |
| 49 | GLM-5.1 Estimated Z.ai (Zhipu) | 22.2% | source | — | — | |
| 50 | MiMo-V2.5-Pro Xiaomi | 22.0% | source | — | — | |
| 51 | GLM-5.3-Flash Supported Z.ai (Zhipu) | 21.0% | source | — | — | |
| 52 | GPT-5.4 mini OpenAI | 21.0% | source | — | — | |
| 53 | Gemini 3 Pro Preview Supported Google | 20.0% | source | — | — | |
| 54 | Claude Sonnet 4.5 (latest) Supported Anthropic | 19.0% | source | — | — | |
| 55 | Beam 1.0 Reflection | 18.0% | source | — | — | |
| 56 | GPT-5 Supported OpenAI | 18.0% | source | — | — | |
| 57 | MiniMax-M3 Estimated MiniMax | 18.0% | source | — | — | |
| 58 | Muse Spark Meta | 17.0% | source | — | — | |
| 59 | DeepSeek V4 Pro Estimated DeepSeek | 16.0% | source | — | — | |
| 60 | Kimi K2.6 Estimated Moonshot AI | 16.0% | source | — | — | |
| 61 | MiMo-V2.5 Xiaomi | 16.0% | source | — | — | |
| 62 | Qwen3.8 27B Alibaba (Qwen) | 16.0% | source | — | — | |
| 63 | Gemini 3 Flash Preview Supported Google | 15.0% | source | — | — | |
| 64 | GPT-5.2 Supported OpenAI | 15.0% | source | — | — | |
| 65 | Grok 4.20 (Non-Reasoning) Estimated xAI | 14.0% | source | — | — | |
| 66 | Gemini 3.5 Flash Lite Google | 13.0% | source | — | — | |
| 67 | GPT-5 Nano OpenAI | 12.0% | source | — | — | |
| 68 | Grok 4.3 Beta xAI | 11.0% | source | — | — | |
| 69 | Mistral Large 4 Mistral | 10.0% | source | — | — | |
| 70 | GPT-5.1 Codex Max Estimated OpenAI | 9.0% | source | — | — | |
| 71 | GPT-5 Mini Estimated OpenAI | 9.0% | source | — | — | |
| 72 | Mistral Medium 3.5 Mistral | 9.0% | source | — | — | |
| 73 | DeepSeek V3.2 Supported DeepSeek | 8.0% | source | — | — | |
| 74 | GLM-4.7 Estimated Z.ai (Zhipu) | 6.0% | source | — | — | |
| 75 | Inkling Small thinkingmachines | 6.0% | source | — | — | |
| 76 | GPT-5.4 nano OpenAI | 5.0% | source | — | — | |
| 77 | Grok 4.1 Fast (Non-Reasoning) xAI | 4.0% | source | — | — | |
| 78 | MiniMax-M2.5 Estimated MiniMax | 4.0% | source | — | — | |
| 79 | Mercury 2.5 inception | 3.0% | source | — | — | |
| 80 | MiniMax-M2.7 MiniMax | 3.0% | source | — | — | |
| 81 | Nemotron 3 Ultra NVIDIA | 2.0% | source | — | — | |
| 82 | Inkling Estimated thinkingmachines | 0.0% | source | — | — | |
| 83 | Laguna M.1 poolside | 0.0% | source | — | — | |
| 84 | Laguna XS.2 Poolside | 0.0% | source | — | — |