ResearchBenchmarksAgentic Reasoning on ZebraArena multi-turn MediumFollow88.14AccuracyCopT74.474478.022281.5785.1178May 19, 2026Evaluation ResultsMethodMethodLinksAccuracyToken CountCopTBackbone=Qwen2.5-35B-A3BBackbone=Qwen2.5-35B-A3B2026.0588.145,457CoTBackbone=Qwen2.5-35B-A3BBackbone=Qwen2.5-35B-A3B2026.05757,217