List your company
ResearchFormerly WizWand

Mathematical Reasoning on AIME (pass@4)

60.43Pass@4ThinkTwice
27.9336.367544.80553.2425Apr 2, 2026
Updated 7mo ago

Evaluation results

MethodLinks
ThinkTwiceBackbone=Qwen3-4B, Training-free=false2026.0460.43
Self-RefineBackbone=Qwen3-4B, Training-free=true2026.0450.37
DAPOBackbone=Qwen3-4B, Training-free=false2026.0449.86
ThinkTwiceBackbone=OLMo3-7B, Training-free=false2026.0449.33
GRPOBackbone=Qwen3-4B, Training-free=false2026.0448.91
Self-RefineBackbone=OLMo3-7B, Training-free=true2026.0447.34
DrGRPOBackbone=Qwen3-4B, Training-free=false2026.0446.98
GRPOBackbone=OLMo3-7B, Training-free=false2026.0446.04
Base ModelBackbone=Qwen3-4B, Training-free=false2026.0445.25
DrGRPOBackbone=OLMo3-7B, Training-free=false2026.0445.24
DAPOBackbone=OLMo3-7B, Training-free=false2026.0444.26
ThinkTwiceBackbone=Qwen3-4B, Evaluation Protocol=Direct prompting, Sampling count (n)=322026.0444.11
DAPOBackbone=Qwen3-4B, Evaluation Protocol=Direct prompting, Sampling count (n)=322026.0442.54
GRPOBackbone=OLMo3-7B, Evaluation Protocol=Direct prompting, Sampling count (n)=322026.0439.38
Base ModelBackbone=OLMo3-7B, Training-free=false2026.0439.31
ThinkTwiceBackbone=OLMo3-7B, Evaluation Protocol=Direct prompting, Sampling count (n)=322026.0439.24
GRPOBackbone=Qwen3-4B, Evaluation Protocol=Direct prompting, Sampling count (n)=322026.0439.06
ReflexionBackbone=Qwen3-4B, Training-free=true2026.0438.47
ReflexionBackbone=OLMo3-7B, Training-free=true2026.0437.38
DAPOBackbone=OLMo3-7B, Evaluation Protocol=Direct prompting, Sampling count (n)=322026.0436.72
DrGRPOBackbone=OLMo3-7B, Evaluation Protocol=Direct prompting, Sampling count (n)=322026.0436.09
DrGRPOBackbone=Qwen3-4B, Evaluation Protocol=Direct prompting, Sampling count (n)=322026.0435.46
Base ModelBackbone=OLMo3-7B, Evaluation Protocol=Direct prompting, Sampling count (n)=322026.0432.81
Base ModelBackbone=Qwen3-4B, Evaluation Protocol=Direct prompting, Sampling count (n)=322026.0429.18