deep-swe

Overview

An open benchmark aimed at measuring how well frontier coding agents handle realistic, long-running engineering work. It matters for teams comparing agent performance, tuning workflows, or publishing reproducible evaluations.