SWE-bench Lite — Eval Explorer
300 instances · UMAP on edit-operation distances · 4 agent models
Color by
Fix type
Repo
Pass/fail
Coverage
Quadrant
Models solved
Filter
All fix types
All repos
Any model outcome
Solved by ≥1 model
Solved by all models
Unsolved by all
Reset
Legend
Pass rate by fix type
Selected instance
Click any point to inspect