List your company

Mathematical Reasoning on GSM (Pass@1, # L)

82.18Pass@1 AccuracySFT-CoT
14.288831.914449.5467.1656May 25, 2026

Evaluation results

MethodLinks
SFT-CoTBackbone=Qwen3-4B2026.0582.1885.2
Ours-RLBackbone=Qwen3-4B2026.0575.2838.36
Ours-SFTBackbone=Qwen3-4B2026.0563.8438.84
SFT-CoTBackbone=Llama-3.2-1B2026.0553.4977.89
Ours-RLBackbone=Llama-3.2-1B2026.0546.4732.09
Ours-SFTBackbone=Llama-3.2-1B2026.0540.7631.25
CoLaR-2Backbone=Qwen3-4B2026.054039.6
SFT-w/o CoTBackbone=Qwen3-4B2026.0538.440
RoTBackbone=Qwen3-4B2026.0537.832
CoconutBackbone=Llama-3.2-1B2026.0524.246
SFT-w/o CoTBackbone=Llama-3.2-1B2026.0521.230
CoLaR-2Backbone=Llama-3.2-1B2026.0517.2337.85
CoconutBackbone=Qwen3-4B2026.0516.96