OpenCompass Reasoning - Puzzle - Chinese (CompassBench 2504): leaderboard
Metric: Score (%). Source: rank.opencompass.org.cn. 28 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | ERNIE-X1-Turbo-32K | 93.7 |
| 2 | Gemini 2.5 Pro (Preview 05-06) | 93.1 |
| 3 | Doubao-1.5-thinking-pro-250415 | 93.1 |
| 4 | DeepSeek R1 | 92.5 |
| 5 | DeepSeek V3 (0324) | 92.5 |
| 6 | O3 Mini (High) | 92.5 |
| 7 | Qwen 3 32B (Thinking) | 92.5 |
| 8 | Qwen 3 235B A22B (Thinking) | 92 |
| 9 | Qwen 3 235B A22B | 90.2 |
| 10 | GPT-4.1 (2025-04-14) | 89.7 |
| 11 | Grok 3 | 89.7 |
| 12 | GLM-4 32B (0414) | 89.1 |
| 13 | O4 Mini (High) | 87.4 |
| 14 | GPT-4.1 Mini | 86.8 |
| 15 | Llama 4 Maverick Instruct | 83.3 |
Interactive version: theaggregate.ai/benchmark?slug=opencompass-reasoning-puzzle-chinese-compassbench-2504 · How It Works · Data refreshed daily, snapshot 2026-09-19.