Reasoning and knowledge
FrontierMath Tiers 1-3
36 models with a published score, best first. Higher is better. The record has been broken 7 times since Apr 16, 2025.
93.7%GPT-6 AstraRank 1, published Sep 3, 2026
Leaderboard
Each model's latest published score. Every score links to its source.
| # | Model | FrontierMath | Date | Note | Source |
|---|---|---|---|---|---|
| 1 | O | 93.7% | Sep 3, 2026 | max setting | epoch.ai |
| 1 | O GPT-6.1 SolOpenAIRun it via OpenAI ChatGPT Plus, OpenAI ChatGPT Pro 100 +3 more | 93.7% | Sep 29, 2026 | max setting | epoch.ai |
| 3 | A Claude Opus 5.5AnthropicRun it via Anthropic Pro, Cursor Premium +5 more | 91.2% | Sep 22, 2026 | max setting | epoch.ai |
| 4 | A Claude Fable 5.1AnthropicRun it via Cursor Premium, Cursor Super +4 more | 90.2% | Sep 1, 2026 | max setting | epoch.ai |
| 5 | O GPT-6 SolOpenAIRun it via OpenAI ChatGPT Plus, OpenAI ChatGPT Pro 100 +3 more | 89.8% | Sep 22, 2026 | max setting | epoch.ai |
| 6 | O GPT-5.6 SolOpenAIRun it via Cursor Premium, Cursor Super +2 more | 89.1% | Jul 9, 2026 | max setting | epoch.ai |
| 7 | A Claude Sonnet 5.5AnthropicRun it via Anthropic Pro, Cursor Premium +5 more | 88.8% | Sep 28, 2026 | max setting | epoch.ai |
| 8 | O GPT-5.5 ProOpenAIRun it via OpenCode Zen | 87.7% | Apr 23, 2026 | xhigh setting | epoch.ai |
| 9 | A Claude Fable 5AnthropicRun it via Cursor Premium, Cursor Super +2 more | 87.0% | Jun 9, 2026 | max setting | epoch.ai |
| 10 | O GPT-5.6 TerraOpenAIRun it via Cursor Premium, Cursor Super +2 more | 86.0% | Jul 9, 2026 | max setting | epoch.ai |
| 10 | G Gemini 4 ArgonGoogleRun it via Google AI Pro, Cursor Premium +5 more | 86.0% | Sep 30, 2026 | high setting | epoch.ai |
| 12 | A Claude Opus 5AnthropicRun it via Cursor Premium, Cursor Super +2 more | 85.6% | Jul 24, 2026 | max setting | epoch.ai |
| 13 | O GPT-5.5OpenAIRun it via Cursor Premium, Cursor Super +2 more | 85.3% | Apr 23, 2026 | xhigh setting | epoch.ai |
| 14 | O GPT-5.4 ProOpenAIRun it via OpenCode Zen | 82.5% | Mar 5, 2026 | xhigh setting | epoch.ai |
| 15 | O GPT-5.6 LunaOpenAIRun it via OpenCode Go, Cursor Premium +4 more | 82.1% | Jul 9, 2026 | max setting | epoch.ai |
| 15 | M Muse WonderMetaRun it via Meta Everyday Usage, Meta High Usage +5 more | 82.1% | Oct 12, 2026 | max setting | epoch.ai |
| 17 | A Claude Opus 4.8AnthropicRun it via Cursor Premium, Cursor Super +2 more | 80.0% | May 28, 2026 | max setting | epoch.ai |
| 18 | O GPT-5.4OpenAIRun it via Cursor Premium, Cursor Super +2 more | 78.6% | Mar 5, 2026 | xhigh setting | epoch.ai |
| 19 | A Claude Haiku 5.5AnthropicRun it via OpenCode Go, Anthropic Pro +7 more | 75.1% | Oct 7, 2026 | max setting | epoch.ai |
| 20 | 74.7% | Aug 2, 2026 | xhigh setting | epoch.ai | |
| 21 | M Muse Spark 1.3MetaRun it via Cursor Premium, Cursor Super +3 more | 74.4% | Sep 2, 2026 | xhigh setting | epoch.ai |
| 22 | M Kimi K3MoonshotRun it via OpenCode Go, Moonshot AI Plus (Kimi Code) +9 more | 72.2% | Jul 16, 2026 | max setting | epoch.ai |
| 23 | G Gemini 3.7 FlashGoogleRun it via Cursor Premium, Cursor Super +2 more | 71.6% | Aug 13, 2026 | high setting | epoch.ai |
| 24 | A Claude Opus 4.7AnthropicRun it via Cursor Premium, Cursor Super +2 more | 70.2% | Apr 16, 2026 | max setting | epoch.ai |
| 25 | M Muse Spark OpenMeta | 69.8% | Oct 15, 2026 | xhigh setting | epoch.ai |
| 26 | 68.8% | Aug 14, 2026 | max setting | epoch.ai | |
| 27 | G Gemini 3.8 FlashGoogleRun it via Cursor Premium, Cursor Super +2 more | 68.4% | Sep 2, 2026 | high setting | epoch.ai |
| 28 | D DeepSeek-V4.1-FlashDeepSeek | 67.4% | Sep 9, 2026 | max setting | epoch.ai |
| 29 | X Grok 4.6xAIRun it via Cursor Plus, Cursor Premium +3 more | 66.0% | Aug 12, 2026 | xhigh setting | epoch.ai |
| 30 | A Claude Sonnet 5AnthropicRun it via Cursor Premium, Cursor Super +2 more | 65.6% | Jun 30, 2026 | max setting | epoch.ai |
| 31 | G Gemini 3.1 ProGoogleRun it via Google AI Pro, Cursor Premium +5 more | 59.6% | Feb 19, 2026 | epoch.ai | |
| 32 | G Gemini 3.6 FlashGoogleRun it via Cursor Premium, Cursor Super +2 more | 59.0% | Jul 21, 2026 | high setting | epoch.ai |
| 33 | M Mistral Large 4MistralRun it via OpenCode Zen | 58.6% | Oct 6, 2026 | high setting | epoch.ai |
| 34 | X Grok 4.5xAIRun it via Cursor Plus, Cursor Premium +3 more | 57.2% | Jul 8, 2026 | high setting | epoch.ai |
| 35 | Z GLM-5.3-FlashZhipuRun it via OpenCode Go, Zhipu GLM Coding Plan Lite +7 more | 55.8% | Aug 20, 2026 | max setting | epoch.ai |
| 36 | X Grok 4.7xAIRun it via Cursor Plus, OpenCode Go +5 more | 53.0% | Sep 21, 2026 | xhigh setting | epoch.ai |
Record history
The best score ever published on FrontierMath Tiers 1-3, and who held it. Each step is a new record.
| Date | Record holder | FrontierMath | Gain | Note |
|---|---|---|---|---|
| Sep 3, 2026 | GPT-6 AstraCurrent | 93.7% | +3.5 | max setting |
| Sep 1, 2026 | Claude Fable 5.1 | 90.2% | +1.1 | max setting |
| Jul 9, 2026 | GPT-5.6 Sol | 89.1% | +1.4 | max setting |
| Apr 23, 2026 | GPT-5.5 Pro | 87.7% | +5.3 | xhigh setting |
| Mar 5, 2026 | GPT-5.4 Pro | 82.5% | +22.8 | xhigh setting |
| Feb 19, 2026 | Gemini 3.1 Pro | 59.6% | +4.2 | |
| Aug 7, 2025 | GPT-5 | 55.4% | +22.1 | high setting |
| Apr 16, 2025 | o3 | 33.3% | First | high setting |
Method
Unpublished expert math problems, run by Epoch AI (CC BY 4.0), best setting per model. Higher is better. Board source: epoch.ai.
We collect the scores the board and the labs publish and run none ourselves. Ranks use each model's latest reading. 40 readings stored for this board, the newest from Oct 15, 2026. The record history includes older models that no longer have a page here.