ResearchBenchmarksMulti-step Reasoning on WordSortFollow100AccuracyBoT96.25697.22898.299.172Jun 1, 2026Evaluation ResultsMethodMethodLinksAccuracyBoTModels=GPT-4Models=GPT-42026.06100eMoTModels=Qwen-32BModels=Qwen-32B2026.0696.8ToTModels=GPT-4Models=GPT-42026.0696.4