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. Saturation forecast: Estimated already saturated. 61 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5 (High) | 99.4 | #98 (GPT-5) |
| 2 | O3 | 99.2 | #124 |
| 3 | GPT-5 (Medium) | 99.13 | #98 (GPT-5) |
| 4 | Grok 4 | 99 | #181 |
| 5 | GPT-5 (Low) | 98.73 | #98 (GPT-5) |
| 6 | Qwen 3 235B A22B 2507 (Thinking) | 98.4 | #260 (Qwen 3 235B A22B 2507) |
| 7 | DeepSeek R1 0528 | 98.27 | #222 |
| 8 | Qwen 3 235B A22B 2507 Instruct | 98 | #300 |
| 9 | GLM-4.5 (Reasoning) | 97.87 | #275 (GLM-4.5) |
| 10 | Qwen 3 30B A3B 2507 (Thinking) | 97.6 | #376 (Qwen 3 30B A3B 2507) |
| 11 | Qwen 3 30B A3B 2507 Instruct | 97.53 | #488 |
| 12 | Kimi K2 | 97.13 | #243 |
| 13 | Gemini 2.5 Pro | 96.73 | #152 |
| 14 | GLM 4.5 Air | 96.53 | #392 |
| 15 | Qwen 3 Coder 480B A35B Instruct | 94.2 | #315 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=aa-math-500 · How It Works · Data refreshed daily, snapshot 2026-10-20.