OpenEval - GPQA CoT: leaderboard
Metric: CoT Correctness (%). Source: huggingface.co. 23 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Qwen 2.5 72B Instruct | 45.07 |
| 2 | Qwen 3 4B Instruct | 39.69 |
| 3 | Qwen 2.5 32B Instruct | 38.79 |
| 4 | Qwen 2 72B Instruct | 37.89 |
| 5 | Qwen 2.5 14B Instruct | 36.1 |
| 6 | Qwen 2.5 7B Instruct Turbo | 34.08 |
| 7 | Qwen 2.5 3B Instruct | 32.74 |
| 8 | Qwen 1.5 32B Chat | 31.39 |
| 9 | Qwen 1.5 14B Chat | 30.72 |
| 10 | Qwen 2.5 7B Instruct | 28.48 |
| 11 | Qwen 2 7B Instruct | 27.8 |
| 12 | Qwen 2.5 1.5B Instruct | 26.68 |
| 13 | Qwen 1.5 7B Chat | 24.44 |
| 14 | Qwen-14B-Chat | 22.2 |
| 15 | Qwen 1.5 4B Chat | 20.85 |
Interactive version: theaggregate.ai/benchmark?slug=openeval-gpqa-cot · How It Works · Data refreshed daily, snapshot 2026-09-05.