List your company

Mathematical Reasoning on MultiArith (pass@1)

82.5Pass@1 RateQwen2.5-1.5B
56.385663.165369.94576.7247Jun 8, 2026

Evaluation results

MethodLinks
Qwen2.5-1.5Bmode=zero-shot2026.0682.5
Coconut + dropout-GRPOseeds=32026.0664.67
Coconut SFTbaseline=true, seeds=12026.0661.67
Coconut + GRPOrollout=deterministic, seeds=32026.0661.67
Coconut + REINFORCE + dropoutbaseline_type=EMA, seeds=32026.0657.39