FrontierMath
Epoch AI leaderboard on unpublished, highly challenging mathematics problems, from undergraduate level (Tiers 1–3) to research level (Tier 4).
At a glance
Claude Fable 5.2 (Max, Default Fallback) leads with 95.1%, out of 128 results. The problems are unpublished and their answers checked automatically, so a model cannot have seen them in training. Each row is one reasoning setting, and two scores whose 95% intervals overlap do not separate.
Data retrieved on Nov 22, 2026.
From undergraduate level to exploratory problems for an advanced graduate student. The home page's Math ranking follows this tier.
- Runs
- 128
- Best score
- 95.1%
- Claude Fable 5.2 (Max, Default Fallback)
- Newest
- Gemini 4 Neon (Xhigh)
- Nov 18, 2026 · 91.2%
- Version
- v2
- Released on 12 June 2026
128 / 128 runs
Released: Nov 9, 2026Std. error: ± 1.3%
Released: Sep 29, 2026Std. error: ± 1.4%
Released: Sep 3, 2026Std. error: ± 1.4%
Released: Nov 17, 2026Std. error: ± 1.6%
Released: Nov 18, 2026Std. error: ± 1.7%
Released: Sep 22, 2026Std. error: ± 1.7%
Released: Oct 26, 2026Std. error: ± 1.7%
Released: Nov 10, 2026Std. error: ± 1.7%
Released: Sep 1, 2026Std. error: ± 1.8%
Released: Sep 22, 2026Std. error: ± 1.8%
Released: Jul 9, 2026Std. error: ± 1.8%
Released: Sep 28, 2026Std. error: ± 1.9%
Released: Nov 12, 2026Std. error: ± 1.9%
Released: Apr 23, 2026Std. error: ± 1.9%
Released: Jun 9, 2026Std. error: ± 2.0%
Released: Oct 13, 2026Std. error: ± 2.1%
Released: Jul 9, 2026Std. error: ± 2.1%
Released: Jul 24, 2026Std. error: ± 2.1%
Released: Apr 23, 2026Std. error: ± 2.1%
Released: Mar 5, 2026Std. error: ± 2.3%
Released: Oct 12, 2026Std. error: ± 2.3%
Released: Jul 9, 2026Std. error: ± 2.3%
Released: May 28, 2026Std. error: ± 2.4%
Released: Sep 22, 2026Std. error: ± 2.4%
Released: Mar 5, 2026Std. error: ± 2.4%
Released: Oct 23, 2026Std. error: ± 2.4%
Released: Oct 7, 2026Std. error: ± 2.6%
Released: Aug 2, 2026Std. error: ± 2.6%
Released: Sep 2, 2026Std. error: ± 2.6%
Released: Sep 2, 2026Std. error: ± 2.6%
- #31GPT-5.2 Proxhigh74.0%
OpenAI
Released: Dec 11, 2025Std. error: ± 2.6%
Released: Jul 16, 2026Std. error: ± 2.7%
Released: Aug 13, 2026Std. error: ± 2.7%
Released: Oct 17, 2026Std. error: ± 2.7%
Released: Apr 16, 2026Std. error: ± 2.7%
Released: Oct 15, 2026Std. error: ± 2.7%
Released: Aug 14, 2026Std. error: ± 2.7%
Released: Sep 2, 2026Std. error: ± 2.8%
Released: Dec 11, 2025Std. error: ± 2.8%
Released: Sep 9, 2026Std. error: ± 2.8%
Released: Aug 12, 2026Std. error: ± 2.8%
Released: Feb 5, 2026Std. error: ± 2.8%
- #43Qwen3.8 Max (0902)xhigh65.6%
Alibaba
Released: Sep 1, 2026Std. error: ± 2.8%
Released: Jun 30, 2026Std. error: ± 2.8%
- #45DeepSeek V4 Pro 0813max64.6%
DeepSeek
Released: Aug 13, 2026Std. error: ± 2.8%
Released: May 19, 2026Std. error: ± 2.8%
Released: May 19, 2026Std. error: ± 2.9%
Released: Feb 19, 2026Std. error: ± 2.9%
Released: Jun 16, 2026Std. error: ± 3.0%
Released: Jul 21, 2026Std. error: ± 2.9%
Released: Oct 6, 2026Std. error: ± 2.9%
- #52DeepSeek V4 Flash 0731max57.5%
DeepSeek
Released: Jul 31, 2026Std. error: ± 2.9%
Released: Jul 8, 2026Std. error: ± 2.9%
- #54Kimi K2.657.2%
Moonshot
Released: Apr 20, 2026Std. error: ± 2.9%
Released: Aug 20, 2026Std. error: ± 2.9%
Released: Oct 7, 2025Std. error: ± 2.9%
Released: Aug 7, 2025Std. error: ± 2.9%
- #58GLM-5.2low54.7%
Z.ai (Zhipu AI)
Released: Jun 16, 2026Std. error: ± 3.0%
Released: Jun 12, 2026Std. error: ± 3.0%
Released: Sep 21, 2026Std. error: ± 3.0%
Released: Mar 17, 2026Std. error: ± 3.0%
Released: Dec 17, 2025Std. error: ± 3.0%
Released: Aug 7, 2025Std. error: ± 3.0%
Released: Jul 15, 2026Std. error: ± 3.0%
Released: Apr 24, 2026Std. error: ± 3.0%
- #66Grok 4.2044.9%
xAI
Released: Feb 17, 2026Std. error: ± 3.0%
- #67GPT-5.4 Nanohigh44.9%
OpenAI
Released: Mar 17, 2026Std. error: ± 3.0%
Released: Apr 17, 2026Std. error: ± 2.9%
- #69GLM-5.242.5%
Z.ai (Zhipu AI)
Released: Jun 16, 2026Std. error: ± 2.9%
Released: Jul 9, 2026Std. error: ± 2.9%
- #71GPT-5.6 Luna39.6%
OpenAI
Released: Jul 9, 2026Std. error: ± 2.9%
Released: Mar 31, 2026Std. error: ± 2.9%
Released: Aug 7, 2025Std. error: ± 2.9%
- #74GLM-5.136.8%
Z.ai (Zhipu AI)
Released: Apr 7, 2026Std. error: ± 2.9%
Released: Apr 16, 2025Std. error: ± 2.9%
- #76Qwen3.6 27B35.1%
Alibaba
Released: Apr 22, 2026Std. error: ± 2.8%
Released: Jun 2, 2026Std. error: ± 2.8%
- #78Claude Opus 4.534.4%
Anthropic
Released: Nov 24, 2025Std. error: ± 2.8%
- #79Qwen3.6 27B34.0%
Alibaba
Released: Apr 22, 2026Std. error: ± 2.8%
Released: Apr 16, 2025Std. error: ± 2.8%
Released: Jul 15, 2026Std. error: ± 2.8%
Released: Mar 31, 2026Std. error: ± 2.8%
- #83Qwen3.5 397B-A17B31.2%
Alibaba
Released: Feb 13, 2026Std. error: ± 2.7%
Released: Apr 16, 2025Std. error: ± 2.7%
- #85Qwen3.5 397B-A17B29.5%
Alibaba
Released: Feb 13, 2026Std. error: ± 2.7%
Released: Apr 16, 2025Std. error: ± 2.7%
Released: Mar 3, 2026Std. error: ± 2.7%
- #88GPT-5.5 Instant26.3%
OpenAI
Released: May 5, 2026Std. error: ± 2.6%
Released: Jul 21, 2026Std. error: ± 2.6%
- #90GLM-5.124.9%
Z.ai (Zhipu AI)
Released: Apr 7, 2026Std. error: ± 2.6%
- #91GPT-5.4 Minilow24.6%
OpenAI
Released: Mar 17, 2026Std. error: ± 2.6%
Released: Jun 17, 2025Std. error: ± 2.6%
- #93Claude Sonnet 4.523.9%
Anthropic
Released: Sep 29, 2025Std. error: ± 2.5%
Released: Mar 3, 2026Std. error: ± 2.5%
- #95Qwen 3.6 Flash22.5%
Alibaba
Released: Apr 27, 2026Std. error: ± 2.5%
Released: Mar 3, 2026Std. error: ± 2.4%
- #97Qwen 3.6 35B-A3B20.4%
Alibaba
Released: Apr 14, 2026Std. error: ± 2.4%
- #98GPT-5.4 Nanolow20.4%
OpenAI
Released: Mar 17, 2026Std. error: ± 2.4%
Released: Aug 7, 2025Std. error: ± 2.4%
- #100Qwen3.7 Flash19.3%
Alibaba
Released: Jul 27, 2026Std. error: ± 2.3%
- #101Qwen3.7 Flash19.3%
Alibaba
Released: Jul 27, 2026Std. error: ± 2.3%
Released: Apr 16, 2025Std. error: ± 2.3%
Released: Sep 24, 2025Std. error: ± 2.3%
Released: Jan 31, 2025Std. error: ± 2.3%
- #105Qwen 3.5 Flash (hosted 35B-A3B)18.2%
Alibaba
Released: Feb 25, 2026Std. error: ± 2.3%
Released: Aug 7, 2025Std. error: ± 2.3%
- #107GPT-5 minilow18.2%
OpenAI
Released: Aug 7, 2025Std. error: ± 2.3%
- #108Qwen 3.6 35B-A3B17.5%
Alibaba
Released: Apr 14, 2026Std. error: ± 2.3%
- #109Qwen 3.6 Flash17.2%
Alibaba
Released: Apr 27, 2026Std. error: ± 2.2%
- #110GPT-5.4 Mini17.2%
OpenAI
Released: Mar 17, 2026Std. error: ± 2.2%
Released: Apr 16, 2025Std. error: ± 2.2%
Released: Dec 17, 2024Std. error: ± 2.1%
- #113Claude Opus 4.112.6%
Anthropic
Released: Aug 5, 2025Std. error: ± 2.0%
Released: Jan 31, 2025Std. error: ± 1.8%
Released: Dec 17, 2024Std. error: ± 1.8%
- #116Qwen 3.5 Flash (hosted 35B-A3B)9.5%
Alibaba
Released: Feb 25, 2026Std. error: ± 1.7%
Released: Dec 17, 2024Std. error: ± 1.6%
Released: Apr 14, 2025Std. error: ± 1.5%
Released: Apr 14, 2025Std. error: ± 1.4%
Released: Aug 7, 2025Std. error: ± 1.4%
- #121GPT-5 nanolow6.0%
OpenAI
Released: Aug 7, 2025Std. error: ± 1.4%
- #122GPT-5.4 Nano4.6%
OpenAI
Released: Mar 17, 2026Std. error: ± 1.2%
Released: Jan 31, 2025Std. error: ± 1.1%
Released: Aug 7, 2025Std. error: ± 0.8%
Released: Jul 18, 2024Std. error: ± 0.5%
Released: Apr 9, 2024Std. error: ± 0.5%
Released: Aug 6, 2024Std. error: ± 0.4%
- #128GPT-3.5 Turbo (Jan 2024)0.0%
OpenAI
Released: Jan 25, 2024Std. error: ± 0.0%
| # | Model | Score | Std. error | Released |
|---|---|---|---|---|
Claude Fable 5.2max Anthropic | 95.1% | ± 1.3% | Nov 9, 2026 | |
| 2 | GPT-6.1 Solmax OpenAI | 93.7% | ± 1.4% | Sep 29, 2026 |
| 3 | GPT-6 Astramax OpenAI | 93.7% | ± 1.4% | Sep 3, 2026 |
| 4 | GPT-6.2 Astramax OpenAI | 92.6% | ± 1.6% | Nov 17, 2026 |
| 5 | Gemini 4 Neonxhigh | 91.2% | ± 1.7% | Nov 18, 2026 |
| 6 | Claude Opus 5.5max Anthropic | 91.2% | ± 1.7% | Sep 22, 2026 |
| 7 | GPT-6.1 Terramax OpenAI | 90.9% | ± 1.7% | Oct 26, 2026 |
| 8 | Gemini 4 Carbonhigh | 90.5% | ± 1.7% | Nov 10, 2026 |
| 9 | Claude Fable 5.1max Anthropic | 90.2% | ± 1.8% | Sep 1, 2026 |
| 10 | GPT-6 Solmax OpenAI | 89.8% | ± 1.8% | Sep 22, 2026 |
| 11 | GPT-5.6 Solmax OpenAI | 89.1% | ± 1.8% | Jul 9, 2026 |
| 12 | Claude Sonnet 5.5max Anthropic | 88.8% | ± 1.9% | Sep 28, 2026 |
| 13 | Gemini 4 Hydrogenxhigh | 88.1% | ± 1.9% | Nov 12, 2026 |
| 14 | GPT-5.5 Proxhigh OpenAI | 87.7% | ± 1.9% | Apr 23, 2026 |
| 15 | Claude Fable 5max Anthropic | 87.0% | ± 2.0% | Jun 9, 2026 |
| 16 | Gemini 4 Argonhigh | 86.0% | ± 2.1% | Oct 13, 2026 |
| 17 | GPT-5.6 Terramax OpenAI | 86.0% | ± 2.1% | Jul 9, 2026 |
| 18 | Claude Opus 5max Anthropic | 85.6% | ± 2.1% | Jul 24, 2026 |
| 19 | GPT-5.5xhigh OpenAI | 85.3% | ± 2.1% | Apr 23, 2026 |
| 20 | GPT-5.4 Proxhigh OpenAI | 82.5% | ± 2.3% | Mar 5, 2026 |
| 21 | Muse Wondermax Meta | 82.1% | ± 2.3% | Oct 12, 2026 |
| 22 | GPT-5.6 Lunamax OpenAI | 82.1% | ± 2.3% | Jul 9, 2026 |
| 23 | Claude Opus 4.8max Anthropic | 80.0% | ± 2.4% | May 28, 2026 |
| 24 | GPT-6 Lunamax OpenAI | 78.9% | ± 2.4% | Sep 22, 2026 |
| 25 | GPT-5.4xhigh OpenAI | 78.6% | ± 2.4% | Mar 5, 2026 |
| 26 | Kimi K3.1 Promax Moonshot AI | 78.2% | ± 2.4% | Oct 23, 2026 |
| 27 | Claude Haiku 5.5max Anthropic | 75.1% | ± 2.6% | Oct 7, 2026 |
| 28 | Qwen 3.8 Maxxhigh Alibaba | 74.7% | ± 2.6% | Aug 2, 2026 |
| 29 | Muse Spark 1.3xhigh Meta | 74.4% | ± 2.6% | Sep 2, 2026 |
| 30 | Muse Spark 1.3max Meta | 74.0% | ± 2.6% | Sep 2, 2026 |
| 31 | GPT-5.2 Proxhigh OpenAI | 74.0% | ± 2.6% | Dec 11, 2025 |
| 32 | Kimi K3max Moonshot AI | 72.2% | ± 2.7% | Jul 16, 2026 |
| 33 | Gemini 3.7 Flashhigh | 71.6% | ± 2.7% | Aug 13, 2026 |
| 34 | DeepSeek V4.1 Flash 1017max DeepSeek | 71.2% | ± 2.7% | Oct 17, 2026 |
| 35 | Claude Opus 4.7max Anthropic | 70.2% | ± 2.7% | Apr 16, 2026 |
| 36 | Muse Spark Openxhigh Meta | 69.8% | ± 2.7% | Oct 15, 2026 |
| 37 | GLM-5.3max Z.AI | 68.8% | ± 2.7% | Aug 14, 2026 |
| 38 | Gemini 3.8 Flashhigh | 68.4% | ± 2.8% | Sep 2, 2026 |
| 39 | GPT-5.2xhigh OpenAI | 67.4% | ± 2.8% | Dec 11, 2025 |
| 40 | DeepSeek V4.1 Flashmax DeepSeek | 67.4% | ± 2.8% | Sep 9, 2026 |
| 41 | Grok 4.6xhigh SpaceXSI | 66.0% | ± 2.8% | Aug 12, 2026 |
| 42 | Claude Opus 4.6max Anthropic | 66.0% | ± 2.8% | Feb 5, 2026 |
| 43 | Qwen3.8 Max (0902)xhigh Alibaba | 65.6% | ± 2.8% | Sep 1, 2026 |
| 44 | Claude Sonnet 5max Anthropic | 65.6% | ± 2.8% | Jun 30, 2026 |
| 45 | DeepSeek V4 Pro 0813max DeepSeek | 64.6% | ± 2.8% | Aug 13, 2026 |
| 46 | Qwen3.7-Max Alibaba | 64.6% | ± 2.8% | May 19, 2026 |
| 47 | Gemini 3.5 Flashhigh | 62.8% | ± 2.9% | May 19, 2026 |
| 48 | Gemini 3.1 Pro | 59.6% | ± 2.9% | Feb 19, 2026 |
| 49 | GLM-5.2max Z.AI | 59.2% | ± 3.0% | Jun 16, 2026 |
| 50 | Gemini 3.6 Flashhigh | 58.9% | ± 2.9% | Jul 21, 2026 |
| 51 | Mistral Large 4 Previewhigh Mistral | 58.6% | ± 2.9% | Oct 6, 2026 |
| 52 | DeepSeek V4 Flash 0731max DeepSeek | 57.5% | ± 2.9% | Jul 31, 2026 |
| 53 | Grok 4.5high SpaceXSI | 57.2% | ± 2.9% | Jul 8, 2026 |
| 54 | Kimi K2.6 Moonshot | 57.2% | ± 2.9% | Apr 20, 2026 |
| 55 | GLM-5.3-Flashmax Z.AI | 55.8% | ± 2.9% | Aug 20, 2026 |
| 56 | GPT-5 Prohigh OpenAI | 55.8% | ± 2.9% | Oct 7, 2025 |
| 57 | GPT-5high OpenAI | 55.4% | ± 2.9% | Aug 7, 2025 |
| 58 | GLM-5.2low Z.ai (Zhipu AI) | 54.7% | ± 3.0% | Jun 16, 2026 |
| 59 | Kimi K2.7 Code Moonshot AI | 54.0% | ± 3.0% | Jun 12, 2026 |
| 60 | Grok 4.7xhigh SpaceXSI | 53.0% | ± 3.0% | Sep 21, 2026 |
| 61 | GPT-5.4 Minixhigh OpenAI | 51.2% | ± 3.0% | Mar 17, 2026 |
| 62 | Gemini 3 Flash Gemini | 51.2% | ± 3.0% | Dec 17, 2025 |
| 63 | GPT-5 minihigh OpenAI | 46.7% | ± 3.0% | Aug 7, 2025 |
| 64 | Inkling-Smallxhigh Thinking Machines | 46.3% | ± 3.0% | Jul 15, 2026 |
| 65 | DeepSeek-V4-Promax DeepSeek | 45.3% | ± 3.0% | Apr 24, 2026 |
| 66 | Grok 4.20 xAI | 44.9% | ± 3.0% | Feb 17, 2026 |
| 67 | GPT-5.4 Nanohigh OpenAI | 44.9% | ± 3.0% | Mar 17, 2026 |
| 68 | Grok 4.3 Betahigh SpaceXSI | 42.8% | ± 2.9% | Apr 17, 2026 |
| 69 | GLM-5.2 Z.ai (Zhipu AI) | 42.5% | ± 2.9% | Jun 16, 2026 |
| 70 | GPT-5.6 Lunalow OpenAI | 41.4% | ± 2.9% | Jul 9, 2026 |
| 71 | GPT-5.6 Luna OpenAI | 39.6% | ± 2.9% | Jul 9, 2026 |
| 72 | Qwen 3.6 Plus Alibaba | 38.2% | ± 2.9% | Mar 31, 2026 |
| 73 | GPT-5low OpenAI | 37.2% | ± 2.9% | Aug 7, 2025 |
| 74 | GLM-5.1 Z.ai (Zhipu AI) | 36.8% | ± 2.9% | Apr 7, 2026 |
| 75 | o4-minihigh OpenAI | 36.1% | ± 2.9% | Apr 16, 2025 |
| 76 | Qwen3.6 27B Alibaba | 35.1% | ± 2.8% | Apr 22, 2026 |
| 77 | Qwen3.7-Plus Alibaba | 34.4% | ± 2.8% | Jun 2, 2026 |
| 78 | Claude Opus 4.5 Anthropic | 34.4% | ± 2.8% | Nov 24, 2025 |
| 79 | Qwen3.6 27B Alibaba | 34.0% | ± 2.8% | Apr 22, 2026 |
| 80 | o3high OpenAI | 33.3% | ± 2.8% | Apr 16, 2025 |
| 81 | Inklingxhigh Thinking Machines | 33.3% | ± 2.8% | Jul 15, 2026 |
| 82 | Qwen 3.6 Plus Alibaba | 32.3% | ± 2.8% | Mar 31, 2026 |
| 83 | Qwen3.5 397B-A17B Alibaba | 31.2% | ± 2.7% | Feb 13, 2026 |
| 84 | o3medium OpenAI | 29.8% | ± 2.7% | Apr 16, 2025 |
| 85 | Qwen3.5 397B-A17B Alibaba | 29.5% | ± 2.7% | Feb 13, 2026 |
| 86 | o4-minimedium OpenAI | 28.8% | ± 2.7% | Apr 16, 2025 |
| 87 | Gemini 3.1 Flash-Litehigh Gemini | 27.7% | ± 2.7% | Mar 3, 2026 |
| 88 | GPT-5.5 Instant OpenAI | 26.3% | ± 2.6% | May 5, 2026 |
| 89 | Gemini 3.5 Flash-Litehigh | 26.0% | ± 2.6% | Jul 21, 2026 |
| 90 | GLM-5.1 Z.ai (Zhipu AI) | 24.9% | ± 2.6% | Apr 7, 2026 |
| 91 | GPT-5.4 Minilow OpenAI | 24.6% | ± 2.6% | Mar 17, 2026 |
| 92 | 24.6% | ± 2.6% | Jun 17, 2025 | |
| 93 | Claude Sonnet 4.5 Anthropic | 23.9% | ± 2.5% | Sep 29, 2025 |
| 94 | Gemini 3.1 Flash-Litelow Gemini | 22.5% | ± 2.5% | Mar 3, 2026 |
| 95 | Qwen 3.6 Flash Alibaba | 22.5% | ± 2.5% | Apr 27, 2026 |
| 96 | Gemini 3.1 Flash-Liteminimal Gemini | 21.4% | ± 2.4% | Mar 3, 2026 |
| 97 | Qwen 3.6 35B-A3B Alibaba | 20.4% | ± 2.4% | Apr 14, 2026 |
| 98 | GPT-5.4 Nanolow OpenAI | 20.4% | ± 2.4% | Mar 17, 2026 |
| 99 | GPT-5 nanohigh OpenAI | 20.0% | ± 2.4% | Aug 7, 2025 |
| 100 | Qwen3.7 Flash Alibaba | 19.3% | ± 2.3% | Jul 27, 2026 |
| 101 | Qwen3.7 Flash Alibaba | 19.3% | ± 2.3% | Jul 27, 2026 |
| 102 | o3low OpenAI | 19.3% | ± 2.3% | Apr 16, 2025 |
| 103 | Qwen3-Max Alibaba | 18.9% | ± 2.3% | Sep 24, 2025 |
| 104 | o3-minihigh OpenAI | 18.6% | ± 2.3% | Jan 31, 2025 |
| 105 | Qwen 3.5 Flash (hosted 35B-A3B) Alibaba | 18.2% | ± 2.3% | Feb 25, 2026 |
| 106 | GPT-5minimal OpenAI | 18.2% | ± 2.3% | Aug 7, 2025 |
| 107 | GPT-5 minilow OpenAI | 18.2% | ± 2.3% | Aug 7, 2025 |
| 108 | Qwen 3.6 35B-A3B Alibaba | 17.5% | ± 2.3% | Apr 14, 2026 |
| 109 | Qwen 3.6 Flash Alibaba | 17.2% | ± 2.2% | Apr 27, 2026 |
| 110 | GPT-5.4 Mini OpenAI | 17.2% | ± 2.2% | Mar 17, 2026 |
| 111 | o4-minilow OpenAI | 16.1% | ± 2.2% | Apr 16, 2025 |
| 112 | o1high OpenAI | 14.7% | ± 2.1% | Dec 17, 2024 |
| 113 | Claude Opus 4.1 Anthropic | 12.6% | ± 2.0% | Aug 5, 2025 |
| 114 | o3-minimedium OpenAI | 10.5% | ± 1.8% | Jan 31, 2025 |
| 115 | o1medium OpenAI | 10.2% | ± 1.8% | Dec 17, 2024 |
| 116 | Qwen 3.5 Flash (hosted 35B-A3B) Alibaba | 9.5% | ± 1.7% | Feb 25, 2026 |
| 117 | o1low OpenAI | 8.4% | ± 1.6% | Dec 17, 2024 |
| 118 | GPT-4.1 mini OpenAI | 6.7% | ± 1.5% | Apr 14, 2025 |
| 119 | GPT-4.1 OpenAI | 6.0% | ± 1.4% | Apr 14, 2025 |
| 120 | GPT-5 miniminimal OpenAI | 6.0% | ± 1.4% | Aug 7, 2025 |
| 121 | GPT-5 nanolow OpenAI | 6.0% | ± 1.4% | Aug 7, 2025 |
| 122 | GPT-5.4 Nano OpenAI | 4.6% | ± 1.2% | Mar 17, 2026 |
| 123 | o3-minilow OpenAI | 3.9% | ± 1.1% | Jan 31, 2025 |
| 124 | GPT-5 nanominimal OpenAI | 1.8% | ± 0.8% | Aug 7, 2025 |
| 125 | GPT-4o mini OpenAI | 0.7% | ± 0.5% | Jul 18, 2024 |
| 126 | GPT-4 Turbo (Apr 2024) OpenAI | 0.7% | ± 0.5% | Apr 9, 2024 |
| 127 | GPT-4o (Aug 2024) OpenAI | 0.4% | ± 0.4% | Aug 6, 2024 |
| 128 | GPT-3.5 Turbo (Jan 2024) OpenAI | 0.0% | ± 0.0% | Jan 25, 2024 |
Method and versions
FrontierMath is an Epoch AI benchmark of several hundred unpublished mathematics problems. Tiers 1 to 3 range from undergraduate problems to exploratory problems suited to an advanced graduate student; Tier 4 is research-level mathematics. BenchSift shows version 2, released on 12 June 2026, which fixed errors in 42% of problems.
- Scores are on the private v2 sets; earlier versions are not mixed into these results.
- Epoch AI evaluates each model at one or more reasoning settings (low to max), each on its own row.
- The standard error is Epoch AI's; the score bar shows the 95% interval it implies.
- A run is tied to a catalogue page by its model id and setting; a run without a matching page keeps Epoch AI's name.
- Epoch AI states that FrontierMath is supported by OpenAI.
- Epoch AI's data is published under the Creative Commons Attribution 4.0 licence.
Data via Epoch AI, FrontierMath v2 (CC BY 4.0).