ResearchBenchmarksTask#8 on STARFor AI assistantsMarkdownContent indexTask#8 on STARFollow82.13Accuracy (ACC)KGRPO78.023580.0767582.1384.18325Oct 22, 2025Evaluation resultsMethodMethodLinksAccuracy (ACC)CoT ScoreKGRPOBackbone=Qwen2.5-VL-7B...Backbone=Qwen2.5-VL-7B, Settings=S22025.1082.1374.48