The hardest difficulty band of the MATH dataset: competition problems across algebra, geometry, number theory and precalculus.
Reliable symbolic and multi-step arithmetic reasoning.
Largely saturated. Useful mostly as a sanity floor for smaller and older models.
| # | Model | Lab | Score | Measured | Method | Source |
|---|---|---|---|---|---|---|
| 1 | GPT-5 | OpenAI | 98.0% | 2025-10-29 | independent · 2 runs | Epoch AI Benchmarking Hub |
| 2 | o3 | OpenAI | 97.8% | 2025-04-16 | independent | Epoch AI Benchmarking Hub |
| 3 | o4-mini | OpenAI | 97.8% | 2025-04-16 | independent | Epoch AI Benchmarking Hub |
| 4 | Claude Sonnet 4.5 | Anthropic | 97.7% | 2025-10-21 | independent | Epoch AI Benchmarking Hub |
| 5 | GPT-5 mini | OpenAI | 97.3% | 2025-10-30 | independent · 2 runs | Epoch AI Benchmarking Hub |
| 6 | Qwen3-Max | Alibaba | 97.1% | 2025-10-09 | independent | Epoch AI Benchmarking Hub |
| 7 | DeepSeek-R1 (May 2025)open | DeepSeek | 96.6% | 2025-05-29 | independent | Epoch AI Benchmarking Hub |
| 8 | Gemini 2.5 Pro (May 2025) | Google DeepMind | 95.9% | 2025-05-08 | independent | Epoch AI Benchmarking Hub |
| 9 | o3-mini | OpenAI | 95.8% | 2025-02-13 | independent · 2 runs | Epoch AI Benchmarking Hub |
| 10 | Gemini 2.5 Pro (Mar 2025) | Google DeepMind | 95.6% | 2025-05-07 | independent | Epoch AI Benchmarking Hub |
| 11 | GPT-5 nano | OpenAI | 95.1% | 2025-08-20 | independent · 2 runs | Epoch AI Benchmarking Hub |
| 12 | o1 | OpenAI | 94.6% | 2025-02-13 | independent · 2 runs | Epoch AI Benchmarking Hub |
| 13 | DeepSeek-R1open | DeepSeek | 93.1% | 2025-01-31 | independent | Epoch AI Benchmarking Hub |
| 14 | Claude Haiku 4.5 | Anthropic | 91.6% | 2025-10-22 | independent · 2 runs | Epoch AI Benchmarking Hub |
| 15 | DeepSeek-R1-Distill-Llama-70Bopen | DeepSeek | 89.9% | 2025-03-10 | independent | Epoch AI Benchmarking Hub |
| 16 | Grok-3 mini | xAI | 89.5% | 2025-04-10 | independent · 2 runs | Epoch AI Benchmarking Hub |
| 17 | Grok 3 | xAI | 88.7% | 2025-04-10 | independent | Epoch AI Benchmarking Hub |
| 18 | GPT-4.1 mini | OpenAI | 87.3% | 2025-04-14 | independent | Epoch AI Benchmarking Hub |
| 19 | DeepSeek-R1-Distill-Qwen-14Bopen | DeepSeek | 87.1% | 2025-03-11 | independent | Epoch AI Benchmarking Hub |
| 20 | o1-mini | OpenAI | 86.7% | 2025-02-13 | independent · 2 runs | Epoch AI Benchmarking Hub |
| 21 | Claude Opus 4 | Anthropic | 85.0% | 2025-05-22 | independent | Epoch AI Benchmarking Hub |
| 22 | Claude Sonnet 4 | Anthropic | 84.4% | 2025-05-22 | independent | Epoch AI Benchmarking Hub |
| 23 | Claude 3.7 Sonnet | Anthropic | 83.9% | 2025-03-13 | independent · 4 runs | Epoch AI Benchmarking Hub |
| 24 | Gemini 2.0 Pro | Google DeepMind | 83.5% | 2025-02-06 | independent | Epoch AI Benchmarking Hub |
| 25 | GPT-4.1 | OpenAI | 83.0% | 2025-04-14 | independent | Epoch AI Benchmarking Hub |
| 26 | Gemini 2.0 Flash | Google DeepMind | 82.2% | 2025-02-06 | independent | Epoch AI Benchmarking Hub |
| 27 | Mistral Medium 3 | Mistral | 81.6% | 2025-05-07 | independent | Epoch AI Benchmarking Hub |
| 28 | o1-preview | OpenAI | 81.6% | 2025-01-27 | independent | Epoch AI Benchmarking Hub |
| 29 | GPT-4.5 | OpenAI | 78.6% | 2025-02-28 | independent | Epoch AI Benchmarking Hub |
| 30 | DeepSeek-V3 (Mar 2025)open | DeepSeek | 75.5% | 2025-04-01 | independent | Epoch AI Benchmarking Hub |
| 31 | Gemma 3 27Bopen | Google DeepMind | 74.0% | 2025-03-13 | independent | Epoch AI Benchmarking Hub |
| 32 | Llama 4 Maverickopen | Meta AI | 73.0% | 2025-04-08 | independent | Epoch AI Benchmarking Hub |
| 33 | GPT-4.1 nano | OpenAI | 70.0% | 2025-04-14 | independent | Epoch AI Benchmarking Hub |
| 34 | Qwen3 235B-A22Bopen | Alibaba | 68.9% | 2025-06-03 | independent | Epoch AI Benchmarking Hub |
| 35 | Qwen2.5-Max | Alibaba | 67.2% | 2025-04-01 | independent | Epoch AI Benchmarking Hub |
| 36 | Qwen Plus | Alibaba | 65.3% | 2025-04-07 | independent | Epoch AI Benchmarking Hub |
| 37 | DeepSeek-V3open | DeepSeek | 64.9% | 2025-01-27 | independent | Epoch AI Benchmarking Hub |
| 38 | Phi-4open | Microsoft | 64.9% | 2025-01-31 | independent | Epoch AI Benchmarking Hub |
| 39 | Grok-2 | xAI | 63.5% | 2025-02-17 | independent | Epoch AI Benchmarking Hub |
| 40 | Qwen2.5-72Bopen | Alibaba | 63.2% | 2025-01-27 | independent | Epoch AI Benchmarking Hub |
| 41 | Llama 4 Scoutopen | Meta AI | 62.3% | 2025-04-08 | independent | Epoch AI Benchmarking Hub |
| 42 | Claude 3.5 Sonnet (October 2024) | Anthropic | 56.9% | 2025-01-27 | independent | Epoch AI Benchmarking Hub |
| 43 | Qwen-Turbo | Alibaba | 56.2% | 2025-04-07 | independent | Epoch AI Benchmarking Hub |
| 44 | Qwen2.5-32Bopen | Alibaba | 56.1% | 2025-01-30 | independent | Epoch AI Benchmarking Hub |
| 45 | Gemini 1.5 Pro | Google DeepMind | 55.6% | 2025-01-27 | independent · 2 runs | Epoch AI Benchmarking Hub |
| 46 | GPT-4o mini | OpenAI | 52.6% | 2025-01-27 | independent | Epoch AI Benchmarking Hub |
| 47 | Claude 3.5 Sonnet | Anthropic | 51.7% | 2025-01-27 | independent | Epoch AI Benchmarking Hub |
| 48 | GPT-4o | OpenAI | 51.4% | 2025-02-05 | independent · 3 runs | Epoch AI Benchmarking Hub |
| 49 | Llama 3.1-405Bopen | Meta AI | 49.8% | 2025-01-27 | independent | Epoch AI Benchmarking Hub |
| 50 | Mistral Large 2open | Mistral | 47.6% | 2025-02-25 | independent · 2 runs | Epoch AI Benchmarking Hub |
| 51 | Mistral Small 3.1open | Mistral | 46.8% | 2025-03-18 | independent | Epoch AI Benchmarking Hub |
| 52 | GPT-4 Turbo (Apr 2024) | OpenAI | 46.7% | 2025-02-27 | independent | Epoch AI Benchmarking Hub |
| 53 | Claude 3.5 Haiku | Anthropic | 46.4% | 2025-03-12 | independent | Epoch AI Benchmarking Hub |
| 54 | Mistral Small 3open | Mistral | 44.8% | 2025-01-30 | independent | Epoch AI Benchmarking Hub |
| 55 | Gemini 1.5 Flash | Google DeepMind | 43.5% | 2025-01-27 | independent · 2 runs | Epoch AI Benchmarking Hub |
| 56 | Tulu 3 (Tülu 3) 70Bopen | Allen Institute | 42.7% | 2025-01-27 | independent | Epoch AI Benchmarking Hub |
| 57 | Llama 3.3 70Bopen | Meta AI | 41.6% | 2025-01-27 | independent | Epoch AI Benchmarking Hub |
| 58 | Llama 3.2 90Bopen | Meta AI | 39.4% | 2025-01-27 | independent | Epoch AI Benchmarking Hub |
| 59 | Qwen2-72Bopen | Alibaba | 39.1% | 2025-01-27 | independent | Epoch AI Benchmarking Hub |
| 60 | GPT-4 Turbo (Nov 2023) | OpenAI | 37.7% | 2025-01-27 | independent · 2 runs | Epoch AI Benchmarking Hub |
| 61 | Claude 3 Opus | Anthropic | 37.5% | 2025-01-27 | independent | Epoch AI Benchmarking Hub |
| 62 | Llama 3.1-70Bopen | Meta AI | 36.7% | 2025-01-27 | independent | Epoch AI Benchmarking Hub |
| 63 | Gemma 2 27Bopen | Google DeepMind | 27.9% | 2025-01-27 | independent | Epoch AI Benchmarking Hub |
| 64 | WizardLM-2 8x22Bopen | Microsoft | 25.7% | 2025-01-27 | independent | Epoch AI Benchmarking Hub |
| 65 | Yi-1.5-34Bopen | Other | 25.5% | 2025-01-27 | independent | Epoch AI Benchmarking Hub |
| 66 | Mistral Large | Mistral | 24.5% | 2025-01-27 | independent | Epoch AI Benchmarking Hub |
| 67 | Mixtral 8x22Bopen | Mistral | 24.2% | 2025-01-27 | independent | Epoch AI Benchmarking Hub |
| 68 | GPT-4 (Jun 2023) | OpenAI | 23.0% | 2025-01-27 | independent | Epoch AI Benchmarking Hub |
| 69 | Llama 3.1-8Bopen | Meta AI | 22.9% | 2025-01-27 | independent | Epoch AI Benchmarking Hub |
| 70 | Hermes 2 Theta Llama-3 70Bopen | Other | 22.7% | 2025-01-27 | independent | Epoch AI Benchmarking Hub |
| 71 | Llama 3-70Bopen | Meta AI | 22.6% | 2025-01-27 | independent | Epoch AI Benchmarking Hub |
| 72 | Gemma 2 9Bopen | Google DeepMind | 21.0% | 2025-01-27 | independent | Epoch AI Benchmarking Hub |
| 73 | Claude 3 Sonnet | Anthropic | 18.2% | 2025-01-27 | independent | Epoch AI Benchmarking Hub |
| 74 | phi-3-medium 14Bopen | Microsoft | 17.6% | 2025-01-31 | independent | Epoch AI Benchmarking Hub |
| 75 | Claude 3 Haiku | Anthropic | 14.9% | 2025-01-27 | independent | Epoch AI Benchmarking Hub |
| 76 | Ministral 8Bopen | Mistral | 14.9% | 2025-01-27 | independent | Epoch AI Benchmarking Hub |
| 77 | Ministral 3B | Mistral | 14.4% | 2025-01-27 | independent | Epoch AI Benchmarking Hub |
| 78 | GPT-3.5 Turbo | OpenAI | 13.8% | 2025-01-27 | independent · 2 runs | Epoch AI Benchmarking Hub |
| 79 | Claude 2 | Anthropic | 11.7% | 2025-01-27 | independent | Epoch AI Benchmarking Hub |
| 80 | DBRXopen | Other | 11.7% | 2025-01-27 | independent | Epoch AI Benchmarking Hub |
| 81 | Gemini 1.0 Pro | Google DeepMind | 11.2% | 2025-01-27 | independent | Epoch AI Benchmarking Hub |
| 82 | Mistral NeMoopen | Mistral | 10.8% | 2025-01-27 | independent | Epoch AI Benchmarking Hub |
| 83 | Mixtral 8x7Bopen | Mistral | 9.6% | 2025-01-27 | independent · 2 runs | Epoch AI Benchmarking Hub |
| 84 | DeepSeek LLM 67Bopen | DeepSeek | 6.4% | 2025-01-27 | independent | Epoch AI Benchmarking Hub |
| 85 | Llama 3-8Bopen | Meta AI | 6.1% | 2025-01-27 | independent | Epoch AI Benchmarking Hub |
| 86 | Yi-34Bopen | Other | 5.1% | 2025-01-27 | independent | Epoch AI Benchmarking Hub |
| 87 | Mistral 7Bopen | Mistral | 3.6% | 2025-01-27 | independent · 2 runs | Epoch AI Benchmarking Hub |
| 88 | Llama 2-70Bopen | Meta AI | 3.3% | 2025-01-27 | independent | Epoch AI Benchmarking Hub |