ResearchBenchmarksN-Back on N-BackFollow97.6AccuracyTASKPR78.77683.66388.5593.437May 25, 2026Evaluation ResultsMethodMethodLinksAccuracyTASKPRModel=Claude Opus 4.6Model=Claude Opus 4.62026.0597.6COMPACTORModel=Llama 3.3 70BModel=Llama 3.3 70B2026.0587.4HumanModel=HumanModel=Human2026.0586.6MEMPRModel=Llama 3 8BModel=Llama 3 8B2026.0579.5