ResearchBenchmarksVision-language Math Reasoning on MathVisionFollow30.5Avg@32 AccuracyMHPO24.98826.41927.8529.281Mar 14, 2026Evaluation ResultsMethodMethodLinksAvg@32 AccuracyMHPOBackbone=Qwen2.5-VL-7B...Backbone=Qwen2.5-VL-7B-Instruct2026.0330.5DAPOBackbone=Qwen2.5-VL-7B...Backbone=Qwen2.5-VL-7B-Instruct2026.0329.1SAPOBackbone=Qwen2.5-VL-7B...Backbone=Qwen2.5-VL-7B-Instruct2026.0328.6GSPOBackbone=Qwen2.5-VL-7B...Backbone=Qwen2.5-VL-7B-Instruct2026.0327.4GRPOBackbone=Qwen2.5-VL-7B...Backbone=Qwen2.5-VL-7B-Instruct2026.0327.3BaselineBackbone=Qwen2.5-VL-7B...Backbone=Qwen2.5-VL-7B-Instruct2026.0325.2