OpenCompass Math - Reasoning Computation - English (CompassBench 2510): leaderboard
Metric: Score (%). Source: rank.opencompass.org.cn. 27 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | O3 (2025-04-16) (High) | 86.2 |
| 2 | GPT-5 (Thinking) | 86.2 |
| 3 | Qwen 3 235B A22B 2507 (Thinking) | 85 |
| 4 | GPT-OSS-120B (High) | 83.8 |
| 5 | DeepSeek R1 0528 | 81.9 |
| 6 | Seed-OSS-36B-Instruct | 81.2 |
| 7 | O4 Mini (2025-04-16) (High) | 80.6 |
| 8 | GLM-4.6 | 80 |
| 9 | Nanbeige3.5-Pro (Thinking) | 78.8 |
| 10 | Grok 4 | 78.1 |
| 11 | Qwen 3 Next 80B A3B (Thinking) | 78.1 |
| 12 | MiniMax-M2 | 76.9 |
| 13 | Claude Sonnet 4.5 (Thinking) | 76.2 |
| 14 | Hunyuan-T1-20250822 | 73.1 |
| 15 | Gemini 2.5 Pro | 72.5 |
Interactive version: theaggregate.ai/benchmark?slug=opencompass-math-reasoning-computation-english-compassbench-2510 · How It Works · Data refreshed daily, snapshot 2026-09-19.