List your company

Mathematical Reasoning on Math500 (Acc, #L)

53.2AccuracySFT-CoT
6.420818.565430.7142.8546May 25, 2026

Evaluation results

MethodLinks
SFT-CoTBackbone=Qwen3-4B2026.0553.2248.2
Ours-RL (gsm)Backbone=Qwen3-4B2026.0542.6107.77
Ours-SFT (gsm)Backbone=Qwen3-4B2026.0541.2132.29
Ours-SFT (math)Backbone=Qwen3-4B2026.0534.4100.03
SFT-w/o CoTBackbone=Qwen3-4B2026.0529.40
Ours-RL (gsm)Backbone=Llama-3.1-1B2026.0514.690.02
SFT-CoTBackbone=Llama-3.1-1B2026.0514205.83
Ours-SFT (math)Backbone=Llama-3.1-1B2026.0511.6112.32
Ours-SFT (gsm)Backbone=Llama-3.1-1B2026.0510.897.8
SFT-w/o CoTBackbone=Llama-3.1-1B2026.058.220