List your company

Grade-school reasoning on GSM Hard

53.4Pass@1 Success RateSFT-CoT
0.224814.029927.83541.6401Jan 21, 2026

Evaluation results

MethodLinks
SFT-CoTModel=Qwen3-VL-4B-Instruct2026.0153.4191.1
SFT-CoTModel=Qwen3-VL-2B-Instruct2026.0133.1207.2
RoT (Ours)Model=Qwen3-VL-4B-Instruct2026.0114.132
SFT-CoTModel=LLaVa-V1.6-Mistral-7B2026.0112.2209.7
SFT-w/o CoTModel=Qwen3-VL-4B-Instruct2026.019.480
RoT (Ours)Model=Qwen3-VL-2B-Instruct2026.018.6432
SFT-w/o CoTModel=Qwen3-VL-2B-Instruct2026.014.70
RoT (Ours)Model=LLaVa-V1.6-Mistral-7B2026.013.6432
SFT-w/o CoTModel=LLaVa-V1.6-Mistral-7B2026.012.270