S1CogBench probes System One decision models across 24 cognitive tasks

Dengzhe Hou · hf · 2026-11-20

S1CogBench evaluates System One decision models on 24 cognitive tasks across 27 modes, spanning Bayesian inference, horizon exploration, rule switching, working memory, information search and social interpretation, with outputs scored on predicting human choices, executing own actions, and reporting explicit probabilities. Findings: fast decisions often lack evidence-based reasoning and reliability varies widely across models.

Original post →

More from Models

Models channel →