Evaluation & Benchmarks · 2024

Introducing SWE-bench Verified

OpenAI, SWE-bench authors

A 500-task human-validated subset of SWE-bench with a containerized harness, created to remove unsolvable and mis-specified tasks so agentic coding scores became a trustworthy frontier metric.

Editorial record

Plain-language summary

A 500-task human-validated subset of SWE-bench with a containerized harness, created to remove unsolvable and mis-specified tasks so agentic coding scores became a trustworthy frontier metric.

Knowledge graph

Relationships

Descendants

Source record

Provenance

Record ID
P-650
Record created
2026-07-13
Last reviewed
2026-07-14
Record version
2

Citation caveat: Citation metadata is approximate and marked unverified in the source dataset.