ResearchBenchmarksMultiple-choice Question Answering on MMLU-Pro (naive task stream)Follow91Accuracy (Economics)ExpRAG85.887.1588.589.85Jun 1, 2026Evaluation ResultsMethodMethodLinksAccuracy (Economics)AccuracyAvg Accuracy (PG)Avg Accuracy (SG)Accuracy (Engineering)Accuracy (Philosophy)ExpRAGPass=1stPass=1st2026.0691—81.3—7380ExpRAGPass=2ndPass=2nd2026.0691——837880MEMPROBEPass=1stPass=1st2026.0691—82.7—7681MEMPROBEPass=2ndPass=2nd2026.0691——837682ReActPass=1stPass=1st2026.0690—78.7—6680ReActPass=2ndPass=2nd2026.0690——78.76680ReMemPass=2ndPass=2nd2026.0687——81.78177ReMemPass=1stPass=1st2026.0686—79—7477