ResearchBenchmarksMultistep Reasoning on MUSR object (test)Follow68.9AccuracyInduced38.42846.33954.2562.161Jun 2, 2026Evaluation ResultsMethodMethodLinksAccuracyInduced2026.0668.9Program-of-Thoughts2026.0660.4Hand2026.0658.5AWM2026.0653.8ReAct2026.0650CoT2026.0639.6