Mathematical Reasoning on HMMT25 (avg@12 accuracy)
48.1Avg@12 AccuracyOPSD
Evaluation Results
| Method | Links | |
|---|---|---|
| OPSDModel=Qwen3-8B2026.05 | 48.1 | |
| SGSDModel=Qwen3-8B2026.05 | 46.9 | |
| SGSDModel=Qwen3-4B2026.05 | 46.7 | |
| OPSDModel=Qwen3-4B2026.05 | 46.1 | |
| GRPOModel=Qwen3-8B2026.05 | 46.1 | |
| GRPOModel=Qwen3-4B2026.05 | 45.3 | |
| BaseModel=Qwen3-8B2026.05 | 45 | |
| GRPO+SkillModel=Qwen3-8B2026.05 | 45 | |
| GRPO+SkillModel=Qwen3-4B2026.05 | 44.7 | |
| OPSD+SkillModel=Qwen3-4B2026.05 | 44.7 | |
| OPSD+SkillModel=Qwen3-8B2026.05 | 43.9 | |
| BaseModel=Qwen3-4B2026.05 | 43.3 | |
| Base+SkillModel=Qwen3-4B2026.05 | 42.2 | |
| Base+SkillModel=Qwen3-8B2026.05 | 40.8 | |
| SGSDModel=Qwen3-1.7B2026.05 | 29.7 | |
| OPSD+SkillModel=Qwen3-1.7B2026.05 | 27.2 | |
| OPSDModel=Qwen3-1.7B2026.05 | 26.9 | |
| GRPOModel=Qwen3-1.7B2026.05 | 25.8 | |
| GRPO+SkillModel=Qwen3-1.7B2026.05 | 25.8 | |
| BaseModel=Qwen3-1.7B2026.05 | 24.2 | |
| Base+SkillModel=Qwen3-1.7B2026.05 | 20.3 |