CogArena

A benchmark for testing AI agents on behavioral experiments.

Kianté Fernandez · Caitlin Chen · Jialu Zhou · Bartek Sadowski · Anthony C. Miceli · Ian Krajbich

University of California, Los Angeles — Department of Psychology

Accepted at NeurIPS 2026 — Evaluations and Datasets Track

Paper Code Leaderboard Try It Yourself Submit

What is CogArena?

CogArena is a benchmark that tests AI agents on behavioral experiments through a web browser. Tasks are sampled widely across cognitive science—from attention and memory to decision-making, reinforcement learning, and social cognition.

Agents must interpret visual stimuli, provide responses, just like a participant in an experiment might.

—

Agents

—

Evaluations

—

Tasks

Live from the leaderboard

Citation

@inproceedings{cogarena2026,
  title         = {{CogArena}: Benchmarking Multimodal Agents on Interactive Behavioral Experiments},
  author        = {Fernandez, Kiant{\'e} and Chen, Caitlin and Zhou, Jialu and Sadowski, Bartek and Miceli, Anthony C. and Krajbich, Ian},
  booktitle     = {Advances in Neural Information Processing Systems (NeurIPS 2026), Evaluations and Datasets Track},
  year          = {2026},
  eprint        = {2610.36484},
  archivePrefix = {arXiv},
  primaryClass  = {cs.AI},
  url           = {https://arxiv.org/abs/2610.36484}
}