Mathematical Reasoning on OlympiadBench (Pass@1 accuracy)
70.7Pass@1 AccuracyPADD
Evaluation Results
| Method | Links | |
|---|---|---|
| PADDModel Family=Qwen Family2026.06 | 70.7 | |
| RSPOModel Family=Qwen Family2026.06 | 66.2 | |
| Teacher (GRPO)Model Family=Qwen Family2026.06 | 63.8 | |
| GSPOModel Family=Qwen Family2026.06 | 63.6 | |
| BaseModel Family=Qwen Family2026.06 | 62.2 | |
| Online KDModel Family=Qwen Family2026.06 | 59.6 | |
| Vanilla-GRPOModel Family=Qwen Family2026.06 | 58.2 | |
| Teacher (GRPO)Model Family=DeepSeek Family2026.06 | 49.7 | |
| PADDModel Family=DeepSeek Family2026.06 | 49.7 | |
| RSPOModel Family=DeepSeek Family2026.06 | 49.6 | |
| GSPOModel Family=DeepSeek Family2026.06 | 47.9 | |
| Online KDModel Family=DeepSeek Family2026.06 | 43.7 | |
| Vanilla-GRPOModel Family=DeepSeek Family2026.06 | 41.8 | |
| Dense-GRPOModel Family=Qwen Family2026.06 | 40.9 | |
| Dense-GRPOModel Family=DeepSeek Family2026.06 | 35.9 | |
| BaseModel Family=DeepSeek Family2026.06 | 31.7 |