FrontierMath Tiers 1-3
Hundreds of unpublished, expert-written math problems.
Agent Index weight
0%
shown for context only
Models tested
94
run by Epoch AI on a shared harness
Results
| # | Model | Score | Normalised | Harness | Cost/task | Date |
|---|---|---|---|---|---|---|
| 1 | Claude Fable 5.2 Supported Anthropic | 95.1% | source | — | Nov 6, 2026 | |
| 2 | GPT-6.1 Sol Estimated OpenAI | 93.7% | source | — | Sep 29, 2026 | |
| 3 | GPT-6 Astra Supported OpenAI | 93.7% | source | — | Aug 30, 2026 | |
| 4 | GPT-6.2 Astra OpenAI | 92.6% | source | — | Oct 27, 2026 | |
| 5 | Claude Opus 5.5 Supported Anthropic | 91.2% | source | — | Sep 22, 2026 | |
| 6 | GPT-6.1 Terra Estimated OpenAI | 90.9% | source | — | Oct 26, 2026 | |
| 7 | Gemini 4 Carbon Supported Google | 90.5% | source | — | Nov 2, 2026 | |
| 8 | Claude Fable 5.1 Supported Anthropic | 90.2% | source | — | Sep 1, 2026 | |
| 9 | GPT-6 Sol Estimated OpenAI | 89.8% | source | — | Sep 22, 2026 | |
| 10 | GPT-5.6 Sol Supported OpenAI | 89.1% | source | — | Jul 9, 2026 | |
| 11 | Claude Sonnet 5.5 Estimated Anthropic | 88.8% | source | — | Sep 29, 2026 | |
| 12 | Gemini 4 Hydrogen Estimated Google | 88.1% | source | — | Nov 13, 2026 | |
| 13 | GPT-5.5 Pro OpenAI | 87.7% | source | — | Jun 12, 2026 | |
| 14 | Claude Fable 5 Supported Anthropic | 87.0% | source | — | Jun 9, 2026 | |
| 15 | Gemini 4 Argon Supported Google | 86.0% | source | — | Oct 13, 2026 | |
| 16 | GPT-5.6 Terra Estimated OpenAI | 86.0% | source | — | Jul 9, 2026 | |
| 17 | Claude Opus 5 Supported Anthropic | 85.6% | source | — | Jul 24, 2026 | |
| 18 | GPT-5.5 Supported OpenAI | 85.3% | source | — | Jun 11, 2026 | |
| 19 | GPT-5.4 Pro OpenAI | 82.5% | source | — | Jun 13, 2026 | |
| 20 | GPT-5.6 Luna Estimated OpenAI | 82.1% | source | — | Jul 9, 2026 | |
| 21 | Muse Wonder Estimated Meta | 82.1% | source | — | Oct 12, 2026 | |
| 22 | Claude Opus 4.8 Supported Anthropic | 80.0% | source | — | Jun 10, 2026 | |
| 23 | GPT-6 Luna OpenAI | 78.9% | source | — | Sep 22, 2026 | |
| 24 | GPT-5.4 Supported OpenAI | 78.6% | source | — | Jun 11, 2026 | |
| 25 | Kimi K3.1 Pro Moonshot AI | 78.2% | source | — | Nov 6, 2026 | |
| 26 | Claude Haiku 5.5 Estimated Anthropic | 75.1% | source | — | Oct 7, 2026 | |
| 27 | Qwen3.8 Max Preview Estimated Alibaba (Qwen) | 74.7% | source | — | Aug 4, 2026 | |
| 28 | Muse Spark 1.3 Estimated Meta | 74.4% | source | — | Sep 16, 2026 | |
| 29 | GPT-5.2 Pro OpenAI | 74.0% | source | — | Jun 13, 2026 | |
| 30 | Kimi K3 Supported Moonshot AI | 72.2% | source | — | Jul 17, 2026 | |
| 31 | Gemini 3.7 Flash Estimated Google | 71.6% | source | — | Aug 14, 2026 | |
| 32 | DeepSeek V4.1 Flash 1017 Estimated DeepSeek | 71.2% | source | — | Oct 19, 2026 | |
| 33 | Claude Opus 4.7 Supported Anthropic | 70.2% | source | — | Jun 10, 2026 | |
| 34 | Muse Spark Open Meta | 69.8% | source | — | Oct 15, 2026 | |
| 35 | GLM-5.3 Supported Z.ai (Zhipu) | 68.8% | source | — | Aug 25, 2026 | |
| 36 | Gemini 3.8 Flash Supported Google | 68.4% | source | — | Sep 2, 2026 | |
| 37 | GPT-5.2 Supported OpenAI | 67.4% | source | — | Jun 11, 2026 | |
| 38 | DeepSeek V4.1 Flash Estimated DeepSeek | 67.4% | source | — | Oct 8, 2026 | |
| 39 | Claude Opus 4.6 Supported Anthropic | 66.0% | source | — | Jun 11, 2026 | |
| 40 | Grok 4.6 Supported xAI | 66.0% | source | — | Aug 14, 2026 | |
| 41 | Claude Sonnet 5 Supported Anthropic | 65.6% | source | — | Jun 30, 2026 | |
| 42 | Qwen3.8 Max 0902 Alibaba (Qwen) | 65.6% | source | — | Sep 2, 2026 | |
| 43 | DeepSeek V4 Pro 0813 DeepSeek | 64.6% | source | — | Aug 19, 2026 | |
| 44 | Qwen3.7 Max Alibaba (Qwen) | 64.6% | source | — | Jun 13, 2026 | |
| 45 | Gemini 3.5 Flash Supported Google | 62.8% | source | — | Jun 10, 2026 | |
| 46 | Gemini 3.1 Pro Preview Supported Google | 59.6% | source | — | Jun 11, 2026 | |
| 47 | GLM-5.2 Supported Z.ai (Zhipu) | 59.2% | source | — | Jun 19, 2026 | |
| 48 | Gemini 3.6 Flash Estimated Google | 58.9% | source | — | Aug 2, 2026 | |
| 49 | Mistral Large 4 Mistral | 58.6% | source | — | Oct 12, 2026 | |
| 50 | DeepSeek V4 Flash 0731 DeepSeek | 57.5% | source | — | Aug 2, 2026 | |
| 51 | Grok 4.5 Estimated xAI | 57.2% | source | — | Jul 9, 2026 | |
| 52 | Kimi K2.6 Estimated Moonshot AI | 57.2% | source | — | Jun 10, 2026 | |
| 53 | GLM-5.3-Flash Supported Z.ai (Zhipu) | 55.8% | source | — | Aug 27, 2026 | |
| 54 | GPT-5 Pro OpenAI | 55.8% | source | — | Jun 12, 2026 | |
| 55 | GPT-5 Supported OpenAI | 55.4% | source | — | Jun 10, 2026 | |
| 56 | Kimi K2.7 Code Estimated Moonshot AI | 54.0% | source | — | Jun 13, 2026 | |
| 57 | Grok 4.7 Estimated xAI | 53.0% | source | — | Sep 22, 2026 | |
| 58 | Gemini 3 Flash Preview Supported Google | 51.2% | source | — | Jun 11, 2026 | |
| 59 | GPT-5.4 mini OpenAI | 51.2% | source | — | Jun 12, 2026 | |
| 60 | GPT-5 Mini Estimated OpenAI | 46.7% | source | — | Jun 12, 2026 | |
| 61 | Inkling Small thinkingmachines | 46.3% | source | — | Aug 14, 2026 | |
| 62 | DeepSeek V4 Pro Estimated DeepSeek | 45.3% | source | — | Jun 17, 2026 | |
| 63 | GPT-5.4 nano OpenAI | 44.9% | source | — | Jun 12, 2026 | |
| 64 | Grok 4.20 (Non-Reasoning) Estimated xAI | 44.9% | source | — | Jul 13, 2026 | |
| 65 | Grok 4.3 Beta xAI | 42.8% | source | — | Jun 17, 2026 | |
| 66 | Qwen3.6 Plus Estimated Alibaba (Qwen) | 38.2% | source | — | Aug 30, 2026 | |
| 67 | GLM-5.1 Estimated Z.ai (Zhipu) | 36.8% | source | — | Aug 29, 2026 | |
| 68 | o4-mini Estimated OpenAI | 36.1% | source | — | Jun 11, 2026 | |
| 69 | Qwen3.6 27B Alibaba (Qwen) | 35.1% | source | — | Aug 28, 2026 | |
| 70 | Claude Opus 4.5 (latest) Supported Anthropic | 34.4% | source | — | Jun 11, 2026 | |
| 71 | Qwen3.7 Plus Alibaba (Qwen) | 34.4% | source | — | Aug 29, 2026 | |
| 72 | Inkling Estimated thinkingmachines | 33.3% | source | — | Aug 6, 2026 | |
| 73 | o3 Supported OpenAI | 33.3% | source | — | Aug 27, 2026 | |
| 74 | Qwen3.5 397B-A17B Alibaba (Qwen) | 31.2% | source | — | Aug 30, 2026 | |
| 75 | Gemini 3.1 Flash Lite Google | 27.7% | source | — | Aug 30, 2026 | |
| 76 | GPT Chat Latest OpenAI | 26.3% | source | — | Aug 2, 2026 | |
| 77 | Gemini 3.5 Flash Lite Google | 26.0% | source | — | Aug 2, 2026 | |
| 78 | Gemini 2.5 Pro Supported Google | 24.6% | source | — | Jun 11, 2026 | |
| 79 | Claude Sonnet 4.5 (latest) Supported Anthropic | 23.9% | source | — | Jun 11, 2026 | |
| 80 | Qwen3.6 Flash Alibaba (Qwen) | 22.5% | source | — | Aug 29, 2026 | |
| 81 | Qwen3.6 35B-A3B Alibaba (Qwen) | 20.4% | source | — | Aug 30, 2026 | |
| 82 | GPT-5 Nano OpenAI | 20.0% | source | — | Jun 12, 2026 | |
| 83 | Qwen3.7 Flash Alibaba (Qwen) | 19.3% | source | — | Aug 29, 2026 | |
| 84 | Qwen3 Max Estimated Alibaba (Qwen) | 18.9% | source | — | Aug 30, 2026 | |
| 85 | o3-mini Estimated OpenAI | 18.6% | source | — | Jun 11, 2026 | |
| 86 | Qwen3.5 Flash Alibaba (Qwen) | 18.2% | source | — | Aug 28, 2026 | |
| 87 | o1 Estimated OpenAI | 14.7% | source | — | Aug 28, 2026 | |
| 88 | Claude Opus 4.1 (latest) Supported Anthropic | 12.6% | source | — | Jun 11, 2026 | |
| 89 | GPT-4.1 mini OpenAI | 6.7% | source | — | Aug 27, 2026 | |
| 90 | GPT-4.1 OpenAI | 6.0% | source | — | Aug 27, 2026 | |
| 91 | GPT-4 Turbo OpenAI | 0.7% | source | — | Aug 27, 2026 | |
| 92 | GPT-4o mini OpenAI | 0.7% | source | — | Aug 27, 2026 | |
| 93 | GPT-4o (2024-08-06) Supported OpenAI | 0.4% | source | — | Aug 27, 2026 | |
| 94 | GPT-3.5-turbo OpenAI | 0.0% | source | — | Aug 27, 2026 |