ResearchBenchmarksLong-context Question Answering on LongBench 1M v2Follow33.5Easy Difficulty AccuracyMemReread25.727.72529.7531.775May 11, 2026Evaluation ResultsMethodMethodLinksEasy Difficulty AccuracyHard Difficulty AccuracyShort Context AccuracyMedium Context AccuracyLong Context AccuracyAverage AccuracyMemRereadScale=4BScale=4B2026.0533.531.232.2293832.1MemAgentScale=4BScale=4B2026.0531.426.432.227.623.128.3ReMemR1Scale=4BScale=4B2026.052624.431.120.524.125