500 challenging competition-level math problems spanning algebra, geometry, number theory, counting, and probability.
Widely used benchmark with stable historical baselines, which makes trend comparisons easier across model generations; broad topic coverage reduces reliance on a single math subdomain.
Its age means some problems may appear in training data. Scores above 90% may reflect memorization rather than novel problem-solving — compare with AIME or LiveBench Math for a less contamination-prone signal.
Higher is better. Accuracy on competition math. Top models score 85-95%.
Max score: 100
Based on ~500 samples, ~300 input tokens and ~500 output tokens per sample (800 total per sample).
Estimates from MATH-500 paper. Actual costs vary with reasoning effort and multi-turn prompting.
Its age means some problems may appear in training data.
| # | Model | Score | Est. Eval Cost | Score/$ |
|---|---|---|---|---|
| 1 | 99.4 | $2.69 | 37.0 | |
| 2 | 99.2 | $2.30 | 43.1 | |
| 3 | 99.2 | $0.17 | 584 | |
| 4 | 99.1 | $2.69 | 36.9 | |
| 5 | 99.1 | $0.0000 | — | |
| 6 | 99.0 | $4.20 | 23.6 | |
| 7 | 98.9 | $1.27 | 78.2 | |
| 8 | 98.7 | $2.69 | 36.7 | |
| 9 | 98.6 | $2.69 | 36.7 | |
| 10 | 98.5 | $1.27 | 77.8 | |
| 11 | 98.4 | $0.61 | 161 | |
| 12 | 98.3 | $0.0000 | — | |
| 13 | 98.3 | $0.97 | 102 | |
| 14 | 98.2 | $21.00 | 4.68 | |
| 15 | 98.1 | $0.67 | 146 | |
| 16 | 98.1 | $0.0000 | — | |
| 17 | 98.0 | $0.25 | 387 | |
| 18 | 98.0 | $0.0000 | — | |
| 19 | 98.0 | $0.63 | 155 | |
| 20 | 97.9 | $0.44 | 223 | |
| 21 | 97.9 | $0.0000 | — | |
| 22 | 97.7 | $0.0000 | — | |
| 23 | 97.6 | $0.0000 | — | |
| 24 | 97.5 | $0.09 | 1.1k | |
| 25 | 97.5 | $0.09 | 1.0k | |
| 26 | 97.3 | $1.27 | 76.9 | |
| 27 | 97.2 | $0.0000 | — | |
| 28 | 97.1 | $0.66 | 147 | |
| 29 | 97.0 | $17.25 | 5.62 | |
| 30 | 96.9 | $0.12 | 842 | |
| 31 | 96.7 | $2.69 | 36.0 | |
| 32 | 96.7 | $0.0000 | — | |
| 33 | 96.6 | $1.30 | 74.3 | |
| 34 | 96.5 | $0.27 | 357 | |
| 35 | 96.3 | $0.0000 | — | |
| 36 | 96.1 | $0.08 | 1.2k | |
| 37 | 96.1 | $0.17 | 562 | |
| 38 | 95.9 | $0.16 | 608 | |
| 39 | 95.9 | $0.0000 | — | |
| 40 | 95.7 | $2.30 | 41.6 | |
| 41 | 95.7 | $0.0000 | — | |
| 42 | 95.4 | $0.0000 | — | |
| 43 | 95.2 | $0.0000 | — | |
| 44 | 94.9 | $0.0000 | — | |
| 45 | 94.7 | $0.0000 | — | |
| 46 | 94.7 | $0.0000 | — | |
| 47 | Gemini 2.0 Flash Thinking Experimental (Jan '25) source artificial_analysis Evidence observed Nov 5, 2026 | 94.4 | $0.0000 | — |
| 48 | 94.4 | $0.0000 | — | |
| 49 | 94.2 | $0.41 | 232 | |
| 50 | 94.2 | $0.26 | 361 |