CogBattery Evaluates Cognitive Task Behavior in 54 LLMs
October 29, 2026
CogBattery benchmarks 54 open-weight models against 13 human cognitive paradigms. Results show that while models respond to n-back lures, they produce significantly more false alarms on non-targets and show weaker confidence-accuracy separation than humans.
HOW THIS AFFECTS YOU
●
researcherYou can use these human-referenced benchmarks to see where model reasoning diverges from human cognitive patterns.