Benchmarks · math
MATH Level 5 (Epoch AI)
The hardest tier (level 5) of the MATH competition set, run by Epoch AI. Accuracy. Data: Epoch AI, CC BY.
61 models · 68 results · data as of 2026-10-29 · epoch.ai ↗
Leaderboard
GPT-5 high
All results
68 rows| # | Model | Effort | MATH Level 5 (Epoch AI) | Harness |
|---|---|---|---|---|
| 1 | GPT-5 | high | 98.1% | Epoch AI |
| 2 | GPT-5 | medium | 97.9% | Epoch AI |
| 3 | GPT-5 Mini | high | 97.8% | Epoch AI |
| 4 | o4 mini | high | 97.8% | Epoch AI |
| 5 | o3 | high | 97.8% | Epoch AI |
| 6 | Sonnet 4.5 | high | 97.7% | Epoch AI |
| 7 | Qwen 3 Max | max | 97.1% | Epoch AI |
| 8 | GPT-5 Mini | medium | 96.8% | Epoch AI |
| 9 | DeepSeek R1 | xhigh | 96.6% | Epoch AI |
| 10 | o3 mini | high | 96.5% | Epoch AI |
| 11 | Haiku 4.5 | xhigh | 96.4% | Epoch AI |
| 12 | Gemini 2.5 Pro | high | 95.9% | Epoch AI |
| 13 | GPT-5 Nano | medium | 95.2% | Epoch AI |
| 14 | o3 mini | medium | 95.2% | Epoch AI |
| 15 | GPT-5 Nano | high | 94.9% | Epoch AI |
| 16 | o1 | high | 94.7% | Epoch AI |
| 17 | o1 | medium | 94.4% | Epoch AI |
| 18 | Grok 3 Mini | low | 90.9% | Epoch AI |
| 19 | o1 mini | high | 89.2% | Epoch AI |
| 20 | Grok 3 | none | 88.7% | Epoch AI |
| 21 | Grok 3 Mini | high | 88.1% | Epoch AI |
| 22 | GPT-4.1 Mini | high | 87.3% | Epoch AI |
| 23 | Opus 4 | medium | 85.0% | Epoch AI |
| 24 | Sonnet 4 | high | 84.4% | Epoch AI |
| 25 | o1 mini | medium | 84.3% | Epoch AI |
| 26 | Gemini 2.0 Pro | none | 83.5% | Epoch AI |
| 27 | GPT-4.1 | high | 83.0% | Epoch AI |
| 28 | Gemini 2.0 Flash | medium | 82.2% | Epoch AI |
| 29 | Mistral Medium 3 | none | 81.6% | Epoch AI |
| 30 | GPT-4.5 | 78.6% | Epoch AI |