AA MATH-500: leaderboard
Artificial Analysis independent evaluation of MATH-500: 500 competition-level math problems across algebra, geometry, number theory, and more.
Metric: Accuracy (%). Source: artificialanalysis.ai. Status: saturated. 61 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Best Expert | 100 |
| 2 | GPT-5 (High) | 99.4 |
| 3 | O3 | 99.2 |
| 4 | GPT-5 (Medium) | 99.13 |
| 5 | Grok 4 | 99 |
| 6 | GPT-5 (Low) | 98.73 |
| 7 | Qwen 3 235B A22B 2507 (Thinking) | 98.4 |
| 8 | DeepSeek R1 0528 | 98.27 |
| 9 | Qwen 3 235B A22B 2507 Instruct | 98 |
| 10 | GLM-4.5 (Reasoning) | 97.87 |
| 11 | Qwen 3 30B A3B 2507 (Thinking) | 97.6 |
| 12 | Qwen 3 30B A3B 2507 Instruct | 97.53 |
| 13 | Kimi K2 | 97.13 |
| 14 | Gemini 2.5 Pro | 96.73 |
| 15 | GLM 4.5 Air | 96.53 |
Interactive version: theaggregate.ai/benchmark?slug=aa-math-500 · How It Works · Data refreshed daily, snapshot 2026-09-05.