FrontierMath measures research-grade mathematical reasoning on curated problems spanning advanced undergraduate to open-problem difficulty. It is designed as a frontier discriminator when standard math sets saturate.
Expert-authored problem set keeps difficulty high across tiers; automated answer-checking scripts reduce subjective grading noise.
Very low absolute solve rates make small improvements statistically unstable; private evaluation process limits external reproducibility and independent auditing.
Higher is better. Accuracy percentage across all tiers.
Max score: 100
Very low absolute solve rates make small improvements statistically unstable.
| # | Model | Score |
|---|---|---|
| 1 | 97.4 | |
| 2 | 95.0 | |
| 3 | 92.1 | |
| 4 | 63.3 | |
| 5 | 63.2 | |
| 6 | 61.6 | |
| 7 | 57.7 | |
| 8 | 57.7 | |
| 9 | 56.4 | |
| 10 | 54.2 | |
| 11 | 54.2 | |
| 12 | 53.8 | |
| 13 | 53.6 | |
| 14 | 53.5 | |
| 15 | 52.7 | |
| 16 | 52.1 | |
| 17 | 51.8 | |
| 18 | 51.4 | |
| 19 | 51.1 | |
| 20 | 51.1 | |
| 21 | 51.0 | |
| 22 | 51.0 | |
| 23 | 50.1 | |
| 24 | 50.0 | |
| 25 | 50.0 | |
| 26 | 49.6 | |
| 27 | 49.6 | |
| 28 | 49.2 | |
| 29 | 49.1 | |
| 30 | 48.9 | |
| 31 | 48.8 | |
| 32 | 47.6 | |
| 33 | 47.5 | |
| 34 | 47.5 | |
| 35 | 47.4 | |
| 36 | 47.2 | |
| 37 | 47.0 | |
| 38 | 46.9 | |
| 39 | 46.1 | |
| 40 | 46.1 | |
| 41 | 45.9 | |
| 42 | 45.4 | |
| 43 | 45.1 | |
| 44 | 44.6 | |
| 45 | 44.3 | |
| 46 | 44.2 | |
| 47 | 43.9 | |
| 48 | 43.8 | |
| 49 | 43.6 | |
| 50 | 43.5 |