ResearchBenchmarksAgent Task Completion on AppWorld (Resampled)Follow66.1Accuracy (%)Metis54.34857.39960.4563.501Jun 23, 2026Evaluation ResultsMethodMethodLinksAccuracy (%)Execution Tokens (K)ReAct TurnsReflection Cost (M)Metis2026.0666.178.510.3211.8SkillX2026.0665.589.511.0317.8ACE2026.0657.7204.811.94.3No Memory2026.0654.8101.713.920