GitHub Launches ReviewBench, an Open Benchmark for Evaluating AI Code Reviewers

GitHub has released ReviewBench, a new offline benchmark that tests AI code‑review agents on over 100 million real GitHub pull requests. The benchmark provides a validated golden set, a consistent scoring rubric, and metrics such as precision, recall, and F1. It enables teams to compare how different reviewers surface issues, balance noise versus coverage, and predict real‑world performance. The benchmark is publicly available for developers to evaluate their own code‑review systems.