FrontierMath Tier 4
Research-level math problems that take experts days.
Agent Index weight
0%
shown for context only
Models tested
76
run by Epoch AI on a shared harness
Results
| # | Model | Score | Normalised | Harness | Cost/task | Date |
|---|---|---|---|---|---|---|
| 1 | Claude Fable 5.2 Supported Anthropic | 100.0% | source | — | Nov 6, 2026 | |
| 2 | GPT-6.1 Sol Estimated OpenAI | 100.0% | source | — | Sep 29, 2026 | |
| 3 | GPT-6.2 Astra OpenAI | 100.0% | source | — | Oct 27, 2026 | |
| 4 | GPT-6 Astra Supported OpenAI | 97.6% | source | — | Aug 30, 2026 | |
| 5 | Gemini 4 Carbon Supported Google | 95.1% | source | — | Nov 2, 2026 | |
| 6 | Claude Opus 5.5 Supported Anthropic | 95.0% | source | — | Sep 22, 2026 | |
| 7 | GPT-6.1 Terra Estimated OpenAI | 92.7% | source | — | Oct 26, 2026 | |
| 8 | Claude Fable 5 Supported Anthropic | 90.2% | source | — | Jun 9, 2026 | |
| 9 | GPT-6 Sol Estimated OpenAI | 90.0% | source | — | Sep 22, 2026 | |
| 10 | Claude Fable 5.1 Supported Anthropic | 87.8% | source | — | Sep 1, 2026 | |
| 11 | Gemini 4 Hydrogen Estimated Google | 87.8% | source | — | Nov 13, 2026 | |
| 12 | GPT-5.6 Sol Supported OpenAI | 82.9% | source | — | Jul 9, 2026 | |
| 13 | Claude Sonnet 5.5 Estimated Anthropic | 80.5% | source | — | Sep 29, 2026 | |
| 14 | Gemini 4 Argon Supported Google | 78.0% | source | — | Oct 13, 2026 | |
| 15 | GPT-5.5 Pro OpenAI | 78.0% | source | — | Jun 12, 2026 | |
| 16 | AI Co-Mathematician Google | 75.6% | source | — | Jun 12, 2026 | |
| 17 | Claude Opus 5 Supported Anthropic | 73.2% | source | — | Jul 24, 2026 | |
| 18 | GPT-5.5 Supported OpenAI | 72.5% | source | — | Jun 11, 2026 | |
| 19 | GPT-5.6 Terra Estimated OpenAI | 70.7% | source | — | Jul 9, 2026 | |
| 20 | Muse Wonder Estimated Meta | 70.7% | source | — | Oct 12, 2026 | |
| 21 | Kimi K3.1 Pro Moonshot AI | 65.9% | source | — | Nov 6, 2026 | |
| 22 | GPT-5.6 Luna Estimated OpenAI | 61.0% | source | — | Jul 9, 2026 | |
| 23 | GPT-5.4 Pro OpenAI | 58.5% | source | — | Jun 13, 2026 | |
| 24 | Claude Opus 4.8 Supported Anthropic | 56.1% | source | — | Jun 10, 2026 | |
| 25 | GPT-6 Luna OpenAI | 56.1% | source | — | Sep 22, 2026 | |
| 26 | GPT-5.4 Supported OpenAI | 49.0% | source | — | Jun 11, 2026 | |
| 27 | Claude Haiku 5.5 Estimated Anthropic | 46.3% | source | — | Oct 7, 2026 | |
| 28 | DeepSeek V4.1 Flash 1017 Estimated DeepSeek | 46.3% | source | — | Oct 19, 2026 | |
| 29 | Muse Spark 1.3 Estimated Meta | 46.3% | source | — | Sep 18, 2026 | |
| 30 | Qwen3.8 Max Preview Estimated Alibaba (Qwen) | 46.3% | source | — | Aug 4, 2026 | |
| 31 | GPT-5.2 Pro OpenAI | 46.0% | source | — | Jun 13, 2026 | |
| 32 | Kimi K3 Supported Moonshot AI | 39.0% | source | — | Jul 17, 2026 | |
| 33 | Gemini 3.7 Flash Estimated Google | 36.6% | source | — | Aug 14, 2026 | |
| 34 | Qwen3.7 Max Alibaba (Qwen) | 34.1% | source | — | Jun 13, 2026 | |
| 35 | Qwen3.8 Max 0902 Alibaba (Qwen) | 34.1% | source | — | Sep 2, 2026 | |
| 36 | Claude Opus 4.7 Supported Anthropic | 31.7% | source | — | Jun 10, 2026 | |
| 37 | Grok 4.6 Supported xAI | 31.7% | source | — | Aug 14, 2026 | |
| 38 | Muse Spark Open Meta | 31.7% | source | — | Oct 15, 2026 | |
| 39 | GPT-5.2 Supported OpenAI | 31.7% | source | — | Jun 11, 2026 | |
| 40 | Claude Sonnet 5 Supported Anthropic | 29.3% | source | — | Jun 30, 2026 | |
| 41 | GLM-5.2 Supported Z.ai (Zhipu) | 29.3% | source | — | Jun 19, 2026 | |
| 42 | GLM-5.3 Supported Z.ai (Zhipu) | 29.3% | source | — | Aug 25, 2026 | |
| 43 | Claude Opus 4.6 Supported Anthropic | 26.8% | source | — | Jun 11, 2026 | |
| 44 | DeepSeek V4.1 Flash Estimated DeepSeek | 26.8% | source | — | Oct 8, 2026 | |
| 45 | DeepSeek V4 Pro 0813 DeepSeek | 26.8% | source | — | Aug 19, 2026 | |
| 46 | Gemini 3.1 Pro Preview Supported Google | 26.8% | source | — | Jun 11, 2026 | |
| 47 | Gemini 3.5 Flash Supported Google | 26.8% | source | — | Jun 10, 2026 | |
| 48 | Mistral Large 4 Mistral | 26.8% | source | — | Oct 12, 2026 | |
| 49 | Kimi K2.6 Estimated Moonshot AI | 25.6% | source | — | Jun 10, 2026 | |
| 50 | DeepSeek V4 Flash 0731 DeepSeek | 24.4% | source | — | Aug 2, 2026 | |
| 51 | Grok 4.5 Estimated xAI | 24.4% | source | — | Jul 9, 2026 | |
| 52 | Gemini 3.6 Flash Estimated Google | 22.0% | source | — | Aug 2, 2026 | |
| 53 | Gemini 3.8 Flash Supported Google | 22.0% | source | — | Sep 2, 2026 | |
| 54 | GPT-5 Supported OpenAI | 22.0% | source | — | Jun 11, 2026 | |
| 55 | GPT-5 Pro OpenAI | 19.5% | source | — | Jun 12, 2026 | |
| 56 | Gemini 3 Flash Preview Supported Google | 17.1% | source | — | Jun 11, 2026 | |
| 57 | GLM-5.3-Flash Supported Z.ai (Zhipu) | 17.1% | source | — | Aug 27, 2026 | |
| 58 | Grok 4.20 (Non-Reasoning) Estimated xAI | 17.1% | source | — | Jul 13, 2026 | |
| 59 | Grok 4.7 Estimated xAI | 17.1% | source | — | Sep 22, 2026 | |
| 60 | Inkling Small thinkingmachines | 17.1% | source | — | Aug 14, 2026 | |
| 61 | Grok 4.3 Beta xAI | 14.6% | source | — | Jun 17, 2026 | |
| 62 | GPT-5.4 nano OpenAI | 12.2% | source | — | Jun 12, 2026 | |
| 63 | GPT-5 Mini Estimated OpenAI | 12.2% | source | — | Jun 12, 2026 | |
| 64 | Kimi K2.7 Code Estimated Moonshot AI | 12.2% | source | — | Jun 13, 2026 | |
| 65 | GPT-5.4 mini OpenAI | 9.8% | source | — | Jun 12, 2026 | |
| 66 | Claude Opus 4.5 (latest) Supported Anthropic | 4.9% | source | — | Jun 11, 2026 | |
| 67 | Inkling Estimated thinkingmachines | 4.9% | source | — | Aug 6, 2026 | |
| 68 | o4-mini Estimated OpenAI | 4.9% | source | — | Jun 11, 2026 | |
| 69 | Claude Opus 4.1 (latest) Supported Anthropic | 2.4% | source | — | Jun 11, 2026 | |
| 70 | Claude Sonnet 4.5 (latest) Supported Anthropic | 2.4% | source | — | Jun 11, 2026 | |
| 71 | DeepSeek V4 Pro Estimated DeepSeek | 2.4% | source | — | Jun 17, 2026 | |
| 72 | GPT Chat Latest OpenAI | 2.4% | source | — | Aug 2, 2026 | |
| 73 | GPT-5 Nano OpenAI | 2.4% | source | — | Jun 12, 2026 | |
| 74 | Gemini 2.5 Pro Supported Google | 0.0% | source | — | Jun 11, 2026 | |
| 75 | Gemini 3.5 Flash Lite Google | 0.0% | source | — | Aug 2, 2026 | |
| 76 | o3-mini Estimated OpenAI | 0.0% | source | — | Jun 11, 2026 |