ResearchBenchmarksAgent Task Performance on AgentBoard ScienceWorld Naive StreamFollow66.7Success Rate (1-step)MEMPROBE58.58860.69462.864.906Jun 1, 2026Evaluation ResultsMethodMethodLinksSuccess Rate (1-step)Precision Rate (1-step)Success Rate (2-step)Precision Rate (2-step)MEMPROBE2026.0666.788.666.788.1ReAct2026.0664.487.164.487.1ReMem2026.0659.383.362.987.1ExpRAG2026.0658.98667.887.3