CogGym
CogGym is a scalable, unified framework for systematically comparing models against humans grounded in cognitive science. CogGym introduces a semi-automated pipeline that standardizes diverse experimental paradigms into a model-agnostic format, enabling reproducible and faithful comparison between humans and model cognition at scale.
Experiments
Human - Model Fit Leaderboard
Full 258-experiment benchmark (91 text, 82 image, 85 video); available experiment counts vary by model. Human split-half R² is Spearman–Brown corrected.
RankModelText R2Image R2Video R2
—
Human split-halfHuman baseline
0.93
0.95
0.92
#1
GPT-5.6 SolOpenAI
0.61
0.63
-
#2
Gemini 3.1 ProGoogle
0.58
0.58
0.42
#3
Claude Opus 4.7Anthropic
0.55
0.45
-
#4
GLM 5.2Z.ai
0.53
-
-
Open submissions
Contribute studies
Share a public or accessible link to a study you would like considered for CogGym.