Real-SWE launches benchmark testing AI coding agents on licensed private enterprise codebases with production engineering tasks

Real-SWE is a new benchmark for evaluating frontier AI models on private, real-world enterprise codebases licensed from companies. The tasks reflect actual engineering work, such as billing, tax calculations, and customer migrations, requiring agents to navigate proprietary systems and company-specific conventions.

The benchmark targets a key gap in AI coding evaluation: whether models can handle enterprise software complexity beyond public GitHub-style tasks. If widely adopted, Real-SWE could shift model competition toward practical reliability, business-context understanding, and production-safe code changes.