AI models
FrontierMath (tiers 1 to 3) leaderboard: AI models ranked
85 models ranked. The top score on FrontierMath belongs to Claude Fable 5.2, at 95.1% per Epoch AI, and 4 models are in a statistical tie with it.
FrontierMath (tiers 1 to 3), on the Epoch AI leaderboard (CC BY 4.0, latest version we checked), is a set of unpublished research-level math problems, kept private by Epoch. The score is the share of correct answers. Here are the 85 models in this section with a score on FrontierMath, per Epoch AI. Rank is shown as a range, because scores within the margin are in a statistical tie.
FrontierMath (tiers 1 to 3)
95.1%
Claude Fable 5.2
models
85
01
| Model | % | Source |
|---|---|---|
| Claude Fable 5.2Anthropic · closed | 95.1% 1st of 85 top score | Epoch AI |
| GPT-6 AstraOpenAI · closed | 93.7% between 2nd and 3rd of 85 | Epoch AI |
| GPT-6.1 SolOpenAI · closed | 93.7% between 2nd and 3rd of 85 | Epoch AI |
| GPT-6.2 AstraOpenAI · closed | 92.6% 4th of 85 | Epoch AI |
| Claude Opus 5.5Anthropic · closed | 91.2% between 5th and 8th of 85 | Epoch AI |
| GPT-6.1 TerraOpenAI · closed | 90.9% between 5th and 8th of 85 | Epoch AI |
| Gemini 4 CarbonGoogle · closed | 90.5% between 5th and 9th of 85 | Epoch AI |
| Claude Fable 5.1Anthropic · closed | 90.2% between 5th and 9th of 85 | Epoch AI |
| GPT-6 SolOpenAI · closed | 89.8% between 7th and 11th of 85 | Epoch AI |
| GPT-5.6 SolOpenAI · closed | 89.1% between 9th and 11th of 85 | Epoch AI |
| Claude Sonnet 5.5Anthropic · closed | 88.8% between 9th and 11th of 85 | Epoch AI |
| GPT-5.5 ProOpenAI · closed | 87.7% between 12th and 13th of 85 | Epoch AI |
| Claude Fable 5Anthropic · closed | 87% between 12th and 15th of 85 | Epoch AI |
| Gemini 4 ArgonGoogle · closed | 86% between 13th and 17th of 85 | Epoch AI |
| GPT-5.6 TerraOpenAI · closed | 86% between 13th and 17th of 85 | Epoch AI |
| Claude Opus 5Anthropic · closed | 85.6% between 14th and 17th of 85 | Epoch AI |
| GPT-5.5OpenAI · closed | 85.3% between 14th and 17th of 85 | Epoch AI |
| GPT-5.4 ProOpenAI · closed | 82.5% between 18th and 20th of 85 | Epoch AI |
| GPT-5.6 LunaOpenAI · closed | 82.1% between 18th and 20th of 85 | Epoch AI |
| Muse WonderMeta · closed | 82.1% between 18th and 20th of 85 | Epoch AI |
| Claude Opus 4.8Anthropic · closed | 80% 21st of 85 | Epoch AI |
| GPT-6 LunaOpenAI · closed | 78.9% between 22nd and 24th of 85 | Epoch AI |
| GPT-5.4OpenAI · closed | 78.6% between 22nd and 24th of 85 | Epoch AI |
| Kimi K3.1 ProMoonshot AI · open | 78.2% between 22nd and 24th of 85 | Epoch AI |
| Claude Haiku 5.5Anthropic · closed | 75.1% between 25th and 26th of 85 | Epoch AI |
| Muse Spark 1.3Meta · closed | 74.4% between 25th and 27th of 85 | Epoch AI |
| GPT-5.2 ProOpenAI · closed | 74% between 26th and 27th of 85 | Epoch AI |
| Kimi K3Moonshot AI · open | 72.2% between 28th and 29th of 85 | Epoch AI |
| Gemini 3.7 FlashGoogle · closed | 71.6% between 28th and 29th of 85 | Epoch AI |
| Claude Opus 4.7Anthropic · closed | 70.2% between 30th and 31st of 85 | Epoch AI |
| Muse Spark OpenMeta · open | 69.8% between 30th and 32nd of 85 | Epoch AI |
| GLM-5.3Zhipu AI (Z.ai) · open | 68.8% between 31st and 33rd of 85 | Epoch AI |
| Gemini 3.8 FlashGoogle · closed | 68.4% between 32nd and 35th of 85 | Epoch AI |
| DeepSeek V4.1 FlashDeepSeek · open | 67.4% between 33rd and 35th of 85 | Epoch AI |
| GPT-5.2OpenAI · closed | 67.4% between 33rd and 35th of 85 | Epoch AI |
| Claude Opus 4.6Anthropic · closed | 66% between 36th and 39th of 85 | Epoch AI |
| Grok 4.6xAI · closed | 66% between 36th and 39th of 85 | Epoch AI |
| Claude Sonnet 5Anthropic · closed | 65.6% between 36th and 40th of 85 | Epoch AI |
| Qwen3.8 MaxAlibaba · closed | 65.6% between 36th and 40th of 85 | Epoch AI |
| Qwen3.7 MaxAlibaba · closed | 64.6% between 38th and 40th of 85 | Epoch AI |
| Gemini 3.5 FlashGoogle · closed | 62.8% 41st of 85 | Epoch AI |
| Gemini 3.1 ProGoogle · closed | 59.6% between 42nd and 44th of 85 | Epoch AI |
| GLM-5.2Zhipu AI (Z.ai) · open | 59.2% between 42nd and 44th of 85 | Epoch AI |
| Gemini 3.6 FlashGoogle · closed | 58.9% between 42nd and 44th of 85 | Epoch AI |
| DeepSeek V4 Flash 0731DeepSeek · open | 57.5% between 45th and 47th of 85 | Epoch AI |
| Grok 4.5xAI · closed | 57.2% between 45th and 47th of 85 | Epoch AI |
| Kimi K2.6Moonshot AI · open | 57.2% between 45th and 47th of 85 | Epoch AI |
| GLM-5.3-FlashZhipu AI (Z.ai) · open | 55.8% between 48th and 50th of 85 | Epoch AI |
| GPT-5 ProOpenAI · closed | 55.8% between 48th and 50th of 85 | Epoch AI |
| GPT-5OpenAI · closed | 55.4% between 48th and 50th of 85 | Epoch AI |
| Kimi K2.7 CodeMoonshot AI · open | 54% between 51st and 52nd of 85 | Epoch AI |
| Grok 4.7xAI · closed | 53% between 51st and 52nd of 85 | Epoch AI |
| Gemini 3 FlashGoogle · closed | 51.2% between 53rd and 54th of 85 | Epoch AI |
| GPT-5.4 MiniOpenAI · closed | 51.2% between 53rd and 54th of 85 | Epoch AI |
| GPT-5 MiniOpenAI · closed | 46.7% 55th of 85 | Epoch AI |
| DeepSeek V4 ProDeepSeek · open | 45.3% between 56th and 58th of 85 | Epoch AI |
| GPT-5.4 NanoOpenAI · closed | 44.9% between 56th and 58th of 85 | Epoch AI |
| Grok 4.20xAI · closed | 44.9% between 56th and 58th of 85 | Epoch AI |
| Grok 4.3 BetaxAI · closed | 42.8% 59th of 85 | Epoch AI |
| Qwen3.6 PlusAlibaba · closed | 38.2% 60th of 85 | Epoch AI |
| GLM 5.1Zhipu AI (Z.ai) · open | 36.8% between 61st and 62nd of 85 | Epoch AI |
| o4-miniOpenAI · closed | 36.1% between 61st and 63rd of 85 | Epoch AI |
| Qwen3.6 27BAlibaba · open | 35.1% between 62nd and 65th of 85 | Epoch AI |
| Claude Opus 4.5Anthropic · closed | 34.4% between 63rd and 65th of 85 | Epoch AI |
| Qwen3.7 PlusAlibaba · closed | 34.4% between 63rd and 65th of 85 | Epoch AI |
| o3OpenAI · closed | 33.3% 66th of 85 | Epoch AI |
| Qwen3.5 397B A17BAlibaba · open | 31.2% 67th of 85 | Epoch AI |
| Gemini 3.1 Flash LiteGoogle · closed | 27.7% 68th of 85 | Epoch AI |
| GPT-5.5 InstantOpenAI · closed | 26.3% between 69th and 70th of 85 | Epoch AI |
| Gemini 3.5 Flash LiteGoogle · closed | 26% between 69th and 70th of 85 | Epoch AI |
| Gemini 2.5 ProGoogle · closed | 24.6% between 71st and 72nd of 85 | Epoch AI |
| Claude Sonnet 4.5Anthropic · closed | 23.9% between 71st and 72nd of 85 | Epoch AI |
| Qwen3.6 FlashAlibaba · undisclosed | 22.5% 73rd of 85 | Epoch AI |
| Qwen3.6 35B A3BAlibaba · open | 20.4% between 74th and 75th of 85 | Epoch AI |
| GPT-5 NanoOpenAI · closed | 20% between 74th and 76th of 85 | Epoch AI |
| Qwen3.7 FlashAlibaba · closed | 19.3% between 75th and 78th of 85 | Epoch AI |
| Qwen3-MaxAlibaba · closed | 18.9% between 76th and 79th of 85 | Epoch AI |
| o3-miniOpenAI · closed | 18.6% between 76th and 79th of 85 | Epoch AI |
| Qwen3.5-FlashAlibaba · undisclosed | 18.2% between 77th and 79th of 85 | Epoch AI |
| o1OpenAI · closed | 14.7% 80th of 85 | Epoch AI |
| Claude Opus 4.1Anthropic · closed | 12.6% 81st of 85 | Epoch AI |
| GPT-4.1 MiniOpenAI · closed | 6.7% between 82nd and 83rd of 85 | Epoch AI |
| GPT-4.1OpenAI · closed | 6% between 82nd and 83rd of 85 | Epoch AI |
| GPT-4 Turbo (Apr 2024)OpenAI · closed | 0.7% between 84th and 85th of 85 | Epoch AI |
| GPT-4o-miniOpenAI · closed | 0.7% between 84th and 85th of 85 | Epoch AI |
FAQ
Which model has the top score on FrontierMath?
The top score on FrontierMath belongs to Claude Fable 5.2, at 95.1% per Epoch AI, and 4 models are in a statistical tie with it.
What does FrontierMath measure?
FrontierMath (tiers 1 to 3), on the Epoch AI leaderboard (CC BY 4.0, latest version we checked), is a set of unpublished research-level math problems, kept private by Epoch. The score is the share of correct answers.
What does “statistical tie with the top score” mean?
It means the gap to the top score is within the 95% margin, which is never below 1 point. On FrontierMath, near 90%, that margin is about 4.9 points. That is why rank is shown as a range, «between Xth and Yth of N»: among tied models, the order may be measurement noise.
Which models are in this ranking?
The 85 models in this section with a published score on FrontierMath, per Epoch AI. No version inherits another's score.
How to read
- Benchmark scores are percentages, and higher is better.
- Rank is shown as a range, «between 1st and 5th of 85». Within the range, the gap to neighboring models falls within the benchmark's margin. The «top score» badge marks the model with the benchmark's top score, and the «statistical tie with the top score» badge marks models whose gap to it is within the margin.
- On a saturated benchmark, the table shows the scores with the badge and without a rank order.
Sources
- Epoch AI, CC BY 4.0, latest version we checked.
- OpenRouter Benchmarks, CC BY 4.0, latest version we checked.
- OpenRouter, dados públicos da API; exports de benchmarks e rankings em CC BY 4.0, latest version we checked.
Dates come from the sources: OpenRouter listing or Hugging Face repo creation, which may differ from the official launch. Family and version were grouped by rule from the name.
How to cite
If you use this data, cite the page and the date you accessed it.
Adriano Meirinho. "FrontierMath (tiers 1 to 3) leaderboard: AI models ranked". meirinho.com.br. https://meirinho.com.br/en/ia/modelos/benchmark/frontiermath/. Accessed November 21, 2026.
Third-party data follows the license of each source. The full table is available as CSV and JSON.