ResearchBenchmarksMultimodal Reasoning on MARS-BenchFollow56.28AccuracyCFPOG54.522454.978755.43555.8913Jun 22, 2026Evaluation ResultsMethodMethodLinksAccuracyCFPOGBackbone=Qwen2.5-VL-7B...Backbone=Qwen2.5-VL-7B, Hyperparameters=gamma=0.02 + No Ent2026.0656.28PAPOGBackbone=Qwen2.5-VL-7BBackbone=Qwen2.5-VL-7B2026.0654.59