OpenCompass Subjective - Reasoning (CompassBench 2405): leaderboard
OpenCompass CompassBench 2405 edition (file dated 2024-06-12): subjective open-ended evaluation, reasoning score. CompassBench is OpenCompass's periodic evaluation of chat models on a private question set; each edition is a separate board. Score 0-100, higher is better.
Metric: Score (%). Source: rank.opencompass.org.cn. Saturation forecast: Estimated already saturated. 39 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-4o (2024-05-13) | 61.6 | #337 |
| 2 | GPT-4 Turbo | 56.9 | #417 |
| 3 | Claude 3 Opus | 56.5 | #565 |
| 4 | DeepSeek V2 Chat | 49 | #572 |
| 5 | Yi Large | 47.4 | #566 |
| 6 | GLM-4 | 47 | #647 |
| 7 | Mistral Large | 44.1 | #713 |
| 8 | Qwen 1.5 110B Chat | 43.3 | #693 |
| 9 | Llama 3 70B Instruct | 42.7 | #640 |
| 10 | Moonshot v1 8K | 42.4 | #622 |
| 11 | Qwen 1.5 72B Chat | 40.4 | #782 |
| 12 | Qwen 1.5 32B Chat | 36.9 | #820 |
| 13 | Mixtral 8x22B Instruct | 35.8 | #738 |
| 14 | Yi 1.5 34B Chat | 34.7 | #798 |
| 15 | Qwen 1.5 14B Chat | 34.4 | #976 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=opencompass-subjective-reasoning-compassbench-2405 · How It Works · Data refreshed daily, snapshot 2026-11-05.