DeepSWE is a benchmark for measuring frontier coding agents on original, long-horizon software engineering tasks. It includes 113 tasks across five languages, isolated environments, program-based verifiers and tooling for reproducible evaluations.
1907DeepSWE is a benchmark for measuring frontier coding agents on original, long-horizon software engineering tasks. It includes 113 tasks across five languages, isolated environments, program-based verifiers and tooling for reproducible evaluations.