CogBattery benchmarks 54 open-weight LLMs against humans on 13 cognitive paradigms

Dengzhe Hou · hf · 2026-11-12

Researchers from Tohoku University, Texas A&M and WPI released CogBattery, a human-referenced benchmark putting 54 open-weight LLMs through text adaptations of 13 cognitive-science paradigms (n-back, decision making, executive function) against original human data.

• Key finding: matching the direction of a human effect doesn't mean models make the same errors or share humans' confidence calibration
• Like humans, models respond more to n-back lures, but produce far more false alarms on ordinary non-targets
• Prior studies covered few paradigms/models, so how general these human-vs-model gaps are was previously unclear

Code: github.com/dengzhe-hou/CogBattery

Original post →

More from Models

• Qwen4 is coming: 'qwen4exp' commits spotted on GitHub — kinetic_energy28 · 2026-11-12
• Fable 5.2 delivers 38% more per limit after price cut, still burns quota 2.9x faster than Opus 5.5 — FinanceYF5 · 2026-11-12
• WMIP Probe Shows LLM Cumulative State Tracking Depends Heavily on Output Protocols — Dengzhe Hou · 2026-11-12
• Baidu ERNIE 5.5 Flash Preview: 280B MoE, 1M context, builds playable games from design docs — LearnWithBishal · 2026-11-12
• Community releases experimental diffusion LLM built on LFM2.5-24B-A2B — alexcovo_eth · 2026-11-12
• MAI's No-Synthetic-Data Mandate vs Nemotron's Synth-Heavy Pretraining Shows Real Divergence — stochasticchasm · 2026-11-12

Models channel →