Mathematical Reasoning on MATH 500 (pass@1, pass@32)
90.4Pass@1 RateClip-Cov
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Clip-CovModel Backbone=Qwen3-8B-Base2025.10 | 90.4 | 97.8 | |
| AERModel Backbone=Qwen3-8B-Base2025.10 | 89.4 | 97 | |
| Clip-HigherModel Backbone=Qwen3-8B-Base2025.10 | 88.3 | 97 | |
| Clip-CovModel Backbone=Qwen3-4B-Base2025.10 | 87.8 | 95.9 | |
| AERModel Backbone=Qwen3-4B-Base2025.10 | 86.7 | 96.6 | |
| Ent-AdvModel Backbone=Qwen3-8B-Base2025.10 | 83.5 | 93.4 | |
| GRPOModel Backbone=Qwen3-8B-Base2025.10 | 82 | 94.1 | |
| Clip-HigherModel Backbone=Qwen3-4B-Base2025.10 | 81.1 | 93.5 | |
| KL-CovModel Backbone=Qwen3-8B-Base2025.10 | 80.6 | 96.7 | |
| GRPOModel Backbone=Qwen3-4B-Base2025.10 | 80.5 | 93.7 | |
| KL-CovModel Backbone=Qwen3-4B-Base2025.10 | 80.3 | 95.5 | |
| Ent-AdvModel Backbone=Qwen3-4B-Base2025.10 | 79.2 | 91.3 | |
| BaseModel Backbone=Qwen3-8B-Base2025.10 | 67.2 | 94.2 | |
| BaseModel Backbone=Qwen3-4B-Base2025.10 | 52.2 | 92 |