Mathematical Reasoning on AIME 25 (avg@12 accuracy)
73.1Avg@12 AccuracyOPSD
Evaluation results
| Method | Links | |
|---|---|---|
| OPSDModel=Qwen3-8B2026.05 | 73.1 | |
| SGSDModel=Qwen3-4B2026.05 | 70.8 | |
| SGSDModel=Qwen3-8B2026.05 | 70.6 | |
| BaseModel=Qwen3-4B2026.05 | 69.7 | |
| GRPOModel=Qwen3-8B2026.05 | 69.7 | |
| OPSDModel=Qwen3-4B2026.05 | 69.2 | |
| Base+SkillModel=Qwen3-8B2026.05 | 68.1 | |
| GRPO+SkillModel=Qwen3-8B2026.05 | 66.9 | |
| GRPOModel=Qwen3-4B2026.05 | 66.7 | |
| OPSD+SkillModel=Qwen3-4B2026.05 | 66.7 | |
| BaseModel=Qwen3-8B2026.05 | 66.1 | |
| Base+SkillModel=Qwen3-4B2026.05 | 65.6 | |
| GRPO+SkillModel=Qwen3-4B2026.05 | 65.3 | |
| OPSD+SkillModel=Qwen3-8B2026.05 | 65.3 | |
| SGSDModel=Qwen3-1.7B2026.05 | 43.6 | |
| OPSDModel=Qwen3-1.7B2026.05 | 43.3 | |
| OPSD+SkillModel=Qwen3-1.7B2026.05 | 40.3 | |
| GRPO+SkillModel=Qwen3-1.7B2026.05 | 38.9 | |
| Base+SkillModel=Qwen3-1.7B2026.05 | 38.3 | |
| BaseModel=Qwen3-1.7B2026.05 | 36.9 | |
| GRPOModel=Qwen3-1.7B2026.05 | 36.7 |