ResearchBenchmarksMulti-step Reasoning on Game24Follow100AccuracyeMoT72.9679.988794.02Jun 1, 2026Evaluation ResultsMethodMethodLinksAccuracyeMoTModels=Qwen-32BModels=Qwen-32B2026.06100BoTModels=GPT-4Models=GPT-42026.0682.4ToTModels=GPT-4Models=GPT-42026.0674