The cost of thinking is similar between large reasoning models and humans
Open Access
- 19 November 2025
- journal article
- Published by National Academy of Sciences in Proceedings of the National Academy of Sciences
- Vol. 122 (47)
Abstract
Do neural network models capture the cognitive demands of human reasoning? Across seven reasoning tasks, we show that the length of the chain-of-thought generated by large reasoning models predicts human reaction times both within tasks—tracking item-level difficulty—and across tasks—capturing broader differences in cognitive demands. This model-to-human alignment shows that out-of-the-box reasoning models reflect core features underlying problem and task complexity in human cognition, without requiring any built-in symbolic mechanisms.Keywords
All Related Versions
Funding Information
- HHS | NIH (U01-NS121471)
This publication has 10 references indexed in Scilit:
- The cost of thinking is similar between large reasoning models and humansProceedings of the National Academy of Sciences, 2025
- The cost of thinking is similar between large reasoning models and humansPsyArXiv, 2025
- A foundation model to predict and capture human cognitionNature, 2025
- Language models, like humans, show content effects on reasoning tasksPNAS Nexus, 2024
- Language is primarily a tool for communication rather than thoughtNature, 2024
- ViSpa (Vision Spaces): A computer-vision-based representation system for individual images and concept prototypes, with large-scale evaluation.Psychological Review, 2023
- Using cognitive psychology to understand GPT-3Proceedings of the National Academy of Sciences, 2023
- What Makes Mental Modeling Difficult? Normative Data for the Multidimensional Relational Reasoning TaskFrontiers in Psychology, 2021
- Rethinking Eliminative ConnectionismCognitive Psychology, 1998
- Connectionism and cognitive architecture: A critical analysisCognition, 1988