Artificial Analysis's composite score for mathematical ability, combining MATH-500, AIME, and other math benchmarks.
Aggregates several math benchmarks to reduce variance from any one test set; helps rank models when individual math leaderboards disagree.
Weighting is set externally, so shifts in component emphasis can move rankings without true capability change; composite obscures which specific math domains drive wins or regressions.
Higher is better. Composite math score — use for relative ranking between models.
Weighting is set externally, so shifts in component emphasis can move rankings without true capability change.
| # | Model | Score |
|---|---|---|
| 1 | 73.7 | |
| 2 | 66.3 | |
| 3 | 55.7 | |
| 4 | 39.0 | |
| 5 | 38.0 | |
| 6 | 37.0 | |
| 7 | 36.3 | |
| 8 | 26.0 | |
| 9 | 11.0 | |
| 10 | 7.7 | |
| 11 | 7.7 | |
| 12 | 4.3 | |
| 13 | 4.0 | |
| 14 | 3.3 | |
| 15 | 3.3 | |
| 16 | 2.3 | |
| 17 | 1.7 | |
| 18 | 0.0 |