List your company

Vision-language Math Reasoning on MathVision

30.5Avg@32 AccuracyMHPO
24.98826.41927.8529.281Mar 14, 2026

Evaluation results

MethodLinks
MHPOBackbone=Qwen2.5-VL-7B-Instruct2026.0330.5
DAPOBackbone=Qwen2.5-VL-7B-Instruct2026.0329.1
SAPOBackbone=Qwen2.5-VL-7B-Instruct2026.0328.6
GSPOBackbone=Qwen2.5-VL-7B-Instruct2026.0327.4
GRPOBackbone=Qwen2.5-VL-7B-Instruct2026.0327.3
BaselineBackbone=Qwen2.5-VL-7B-Instruct2026.0325.2