LiveFR

SWE-bench

Definition: SWE-bench is a benchmark that evaluates a model's ability to solve real software engineering issues drawn from open-source repositories.

The model must produce a patch that passes the project's tests, measuring realistic coding skill. It is often used to compare development-focused models.

See also

← Full AI glossary · AI news