The hardest tier from MathArena's 2025 evaluation — research-level mathematics problems that push the boundary of what AI can solve.
This definition has not completed editorial and provenance review. Verify its metric, version, source, and evaluation configuration upstream before using it for a model-selection decision.
Extreme difficulty designed to differentiate the very best mathematical reasoners. Fresh 2025 problems.
Very low scores across all models (max ~34%) — limited differentiation at the lower end.
Higher is better. Accuracy percentage (0–100%). Even top models score <35%.
Very low scores across all models (max ~34%) — limited differentiation at the lower end.
| # | Model | Score |
|---|---|---|
| 1 | 99.8 | |
| 2 | 99.2 | |
| 3 | 98.0 | |
| 4 | 53.2 | |
| 5 | 51.4 | |
| 6 | 51.2 | |
| 7 | 49.7 | |
| 8 | 47.5 | |
| 9 | 40.5 | |
| 10 | 38.0 | |
| 11 | 36.9 | |
| 12 | 36.5 | |
| 13 | 36.4 | |
| 14 | 34.3 | |
| 15 | 33.5 | |
| 16 | 33.5 | |
| 17 | 33.1 | |
| 18 | 32.6 | |
| 19 | 32.1 | |
| 20 | 30.1 | |
| 21 | 29.8 | |
| 22 | 29.8 | |
| 23 | 29.7 | |
| 24 | 27.6 | |
| 25 | 27.4 | |
| 26 | 26.5 | |
| 27 | 23.5 | |
| 28 | 23.4 | |
| 29 | 23.1 | |
| 30 | 22.9 | |
| 31 | 22.5 | |
| 32 | 22.2 | |
| 33 | 22.2 | |
| 34 | 21.0 | |
| 35 | 20.3 | |
| 36 | 20.1 | |
| 37 | 20.1 | |
| 38 | 19.3 | |
| 39 | 19.0 | |
| 40 | 18.2 | |
| 41 | 17.5 | |
| 42 | 16.5 | |
| 43 | 16.1 | |
| 44 | 15.9 | |
| 45 | 15.6 | |
| 46 | 15.6 | |
| 47 | 15.5 | |
| 48 | 15.4 | |
| 49 | 14.3 | |
| 50 | 14.2 |