STEM Reasoning on GPQA Diamond (Accuracy avg@4)
55Accuracy (avg@4)PAPO
Evaluation Results
| Method | Links | |
|---|---|---|
| PAPOModel=Qwen2.5-14B2026.03 | 55 | |
| ORM(GRPO)Model=Qwen2.5-14B2026.03 | 47 | |
| PAPOModel=Qwen3-4B-Base2026.03 | 43 | |
| PAPOModel=Qwen2.5-7B2026.03 | 42.4 | |
| ORM(GRPO)Model=Qwen2.5-7B2026.03 | 40.7 | |
| ORM(DAPO)Model=Qwen3-4B-Base2026.03 | 39 | |
| ORM(GRPO)Model=Qwen2.5-3B2026.03 | 38.8 | |
| PAPOModel=Qwen2.5-3B2026.03 | 36.3 | |
| BaseModel=Qwen3-4B-Base2026.03 | 34.9 | |
| BaseModel=Qwen2.5-14B2026.03 | 34.4 | |
| BaseModel=Qwen2.5-7B2026.03 | 30.1 | |
| BaseModel=Qwen2.5-3B2026.03 | 27.6 |