View on GitHub

Evaluating Decision Models on Visual Complexity Using CAPTCHAs

Synthetic benchmark only (not for bypassing real CAPTCHAs). Per-cell question: “Is a stoplight visible in this image?” · 150 images · 2,460 cell decisions

Accuracy vs Grid Size

Model Comparison

Loading backends…

Leaderboard

image ⇅thumbmodelgrid ⇅diff ⇅cell-acc ⇅full ✓ ⇅avg conf ⇅total ms ⇅