Benchmarks
MATH Level 5
Models scored
108
Best score
0.98
Metric
mean_score
MATH Level 5 is an AI evaluation tracked by Epoch AI, with 108 scored model versions on record. Epoch AI runs this evaluation directly, so the scores are reproducible against their published logs.
Scored models span June 2023 to October 2025. The best score recorded is 0.98, measured as mean_score. Models come from OpenAI, Anthropic, Alibaba, DeepSeek, Google DeepMind, Mistral AI, xAI, Meta AI and others.
Full record
- Score metric
- mean_score
- Models scored
- 108
- Best score recorded
- 0.98
- Earliest model scored
- Jun 13, 2023
- Most recent model scored
- Oct 15, 2025
- Third-party reported
- No
Leaderboard
| 01 | gpt-5-2025-08-07_high | OpenAI | 0.98 |
| 02 | gpt-5-2025-08-07_medium | OpenAI | 0.98 |
| 03 | gpt-5-mini-2025-08-07_high | OpenAI | 0.98 |
| 04 | o4-mini-2025-04-16_high | OpenAI | 0.98 |
| 05 | o3-2025-04-16_high | OpenAI | 0.98 |
| 06 | claude-sonnet-4-5-20250929_32K | Anthropic | 0.98 |
| 07 | qwen3-max-2025-09-23 | Alibaba | 0.97 |
| 08 | gpt-5-mini-2025-08-07_medium | OpenAI | 0.97 |
| 09 | DeepSeek-R1-0528 | DeepSeek | 0.97 |
| 10 | o3-mini-2025-01-31_high | OpenAI | 0.96 |
| 11 | claude-haiku-4-5-20251001_32K | Anthropic | 0.96 |
| 12 | gemini-2.5-pro-preview-05-06 | Google DeepMind | 0.96 |
| 13 | gemini-2.5-pro-preview-03-25 | Google DeepMind | 0.96 |
| 14 | gpt-5-nano-2025-08-07_medium | OpenAI | 0.95 |
| 15 | o3-mini-2025-01-31_medium | OpenAI | 0.95 |
| 16 | gpt-5-nano-2025-08-07_high | OpenAI | 0.95 |
| 17 | o1-2024-12-17_high | OpenAI | 0.95 |
| 18 | o1-2024-12-17_medium | OpenAI | 0.94 |
| 19 | DeepSeek-R1 | DeepSeek | 0.93 |
| 20 | claude-3-7-sonnet-20250219_64K | Anthropic | 0.91 |
| 21 | grok-3-mini-beta_low | xAI | 0.91 |
| 22 | claude-3-7-sonnet-20250219_32K | Anthropic | 0.9 |
| 23 | DeepSeek-R1-Distill-Llama-70B | DeepSeek | 0.9 |
| 24 | o1-mini-2024-09-12_high | OpenAI | 0.89 |
| 25 | grok-3-beta | xAI | 0.89 |
| 26 | grok-3-mini-beta_high | xAI | 0.88 |
| 27 | gpt-4.1-mini-2025-04-14 | OpenAI | 0.87 |
| 28 | DeepSeek-R1-Distill-Qwen-14B | DeepSeek | 0.87 |
| 29 | claude-haiku-4-5-20251001 | Anthropic | 0.87 |
| 30 | claude-3-7-sonnet-20250219_16K | Anthropic | 0.86 |
| 31 | claude-opus-4-20250514 | Anthropic | 0.85 |
| 32 | claude-sonnet-4-20250514 | Anthropic | 0.84 |
| 33 | o1-mini-2024-09-12_medium | OpenAI | 0.84 |
| 34 | gemini-2.0-pro-exp-02-05 | Google DeepMind | 0.83 |
| 35 | gpt-4.1-2025-04-14 | OpenAI | 0.83 |
| 36 | gemini-2.0-flash-001 | Google DeepMind,Google | 0.82 |
| 37 | o1-preview-2024-09-12 | OpenAI | 0.82 |
| 38 | mistral-medium-2505 | Mistral AI | 0.82 |
| 39 | gpt-4.5-preview-2025-02-27 | OpenAI | 0.79 |
| 40 | DeepSeek-V3-0324 | DeepSeek | 0.76 |
| 41 | gemma-3-27b-it | Google DeepMind | 0.74 |
| 42 | Llama-4-Maverick-17B-128E-Instruct-FP8 | Meta AI | 0.73 |
| 43 | gemini-1.5-pro-002 | Google DeepMind | 0.7 |
| 44 | gpt-4.1-nano-2025-04-14 | OpenAI | 0.7 |
| 45 | qwen3-235b-a22b | Alibaba | 0.69 |
| 46 | claude-3-7-sonnet-20250219 | Anthropic | 0.68 |
| 47 | qwen-max-2025-01-25 | Alibaba | 0.67 |
| 48 | qwen-plus-2025-01-25 | Alibaba | 0.65 |
| 49 | phi-4 | Microsoft Research | 0.65 |
| 50 | DeepSeek-V3 | DeepSeek | 0.65 |
| 51 | grok-2-1212 | xAI | 0.64 |
| 52 | qwen2.5-72b-instruct | Alibaba | 0.63 |
| 53 | Llama-4-Scout-17B-16E-Instruct | Meta AI | 0.62 |
| 54 | gemini-1.5-flash-002 | Google DeepMind | 0.62 |
| 55 | claude-3-5-sonnet-20241022 | Anthropic | 0.57 |
| 56 | qwen-turbo-2024-11-01 | Alibaba | 0.56 |
| 57 | qwen2.5-32b-instruct | Alibaba | 0.56 |
| 58 | gpt-4o-2024-08-06 | OpenAI | 0.53 |
| 59 | gpt-4o-mini-2024-07-18 | OpenAI | 0.53 |
| 60 | claude-3-5-sonnet-20240620 | Anthropic | 0.52 |
More benchmarks
SourceEpoch AI, 'AI Benchmarking Hub'. Published online at epoch.ai. Retrieved 2026-07-29 from https://epoch.ai/benchmarks. Licensed under CC BY 4.0.
← All benchmarks