Mathematical Reasoning on MATH 500 (pass@4)
95.7Pass@4 RateThinkTwice
Evaluation Results
| Method | Links | |
|---|---|---|
| ThinkTwiceBackbone=Qwen3-4B, Training-free=false2026.04 | 95.7 | |
| DAPOBackbone=Qwen3-4B, Training-free=false2026.04 | 94.96 | |
| ThinkTwiceBackbone=OLMo3-7B, Training-free=false2026.04 | 94.66 | |
| DrGRPOBackbone=Qwen3-4B, Training-free=false2026.04 | 94.46 | |
| Self-RefineBackbone=Qwen3-4B, Training-free=true2026.04 | 93.86 | |
| GRPOBackbone=Qwen3-4B, Training-free=false2026.04 | 93.78 | |
| DrGRPOBackbone=OLMo3-7B, Training-free=false2026.04 | 93.54 | |
| DAPOBackbone=OLMo3-7B, Training-free=false2026.04 | 93.33 | |
| Self-RefineBackbone=OLMo3-7B, Training-free=true2026.04 | 93.24 | |
| Base ModelBackbone=Qwen3-4B, Training-free=false2026.04 | 92.82 | |
| GRPOBackbone=OLMo3-7B, Training-free=false2026.04 | 92.28 | |
| Base ModelBackbone=OLMo3-7B, Training-free=false2026.04 | 91.75 | |
| ReflexionBackbone=Qwen3-4B, Training-free=true2026.04 | 91.48 | |
| ReflexionBackbone=OLMo3-7B, Training-free=true2026.04 | 91.29 |