Best LLM for math proofs
- Claude Fable 5.1 leads
- 100.0%
- models ranked
- 74
ProofBench · read October 21
As of October 21, 2026, Anthropic Claude Fable 5.1 leads 74 models on ProofBench with 100.0%, ahead of Anthropic Claude Opus 5.5 (100.0%) and Anthropic Claude Sonnet 5.5 (100.0%).
The full best LLM for math ranking combines this benchmark with others.
| Rank | Model | ProofBench | Coding evidence | Coding index | Output $/Mtok |
|---|---|---|---|---|---|
| 1 | Claude Fable 5.1 Anthropic | 100.0% | 84.2 | $50.00 | |
| 2 | Claude Opus 5.5 Anthropic | 100.0% | 88.8 | $20.00 | |
| 3 | Claude Sonnet 5.5 Anthropic | 100.0% | 87.3 | $10.00 | |
| 4 | Claude Opus 5 Anthropic | 99.0% | 83.0 | $25.00 | |
| 5 | Gemini 4 Argon Google | 99.0% | 81.0 | $10.00 | |
| 6 | GPT-6 Astra OpenAI | 99.0% | 85.1 | $50.00 | |
| 7 | GPT-6.1 Sol OpenAI | 99.0% | 82.4 | $10.00 | |
| 8 | Claude Fable 5 Anthropic | 95.0% | 79.3 | $50.00 | |
| 9 | Claude Haiku 5.5 Anthropic | 89.0% | 70.7 | $0.50 | |
| 10 | Kimi K3 Moonshot AI | 87.0% | 62.3 | $15.00 | |
| 11 | Muse Wonder Meta | 84.0% | 71.9 | $20.00 | |
| 12 | GPT-5.6 Sol OpenAI | 83.0% | 73.5 | $20.00 | |
| 13 | GPT-6 Sol OpenAI | 83.0% | 76.0 | $10.00 | |
| 14 | Claude Sonnet 5 Anthropic | 77.0% | 57.7 | $10.00 | |
| 15 | Hy4 preview Tencent | 75.0% | 44.3 | $2.2509 | |
| 16 | GPT-5.6 Terra OpenAI | 74.0% | 63.9 | $12.00 | |
| 17 | MiMo-V2.6-Pro Xiaomi | 70.0% | 65.4 | $0.87 | |
| 18 | Claude Opus 4.8 Anthropic | 69.0% | 65.0 | $25.00 | |
| 19 | GPT-6 Luna OpenAI | 64.0% | 57.3 | $0.50 | |
| 20 | MiMo-V2.6-Flash Xiaomi | 63.0% | 54.7 | $0.28 |
How this list is built
Generated from the dataset on every refresh. Nobody picks the order by hand.
Models by their ProofBench result, as the board publishes it, best first. Epoch AI collects its results; its benchmarking hub describes the tasks and links the authors.
- OpenRouter, fetched
- Epoch AI benchmarking hub, fetched