GitHub launches ReviewBench, an open AI code-review benchmark built from 219 real pull requests across 19 languages

What changed

GitHub released ReviewBench in research preview on October 5. The benchmark contains 219 public pull requests from 187 open-source repositories across 19 languages, with distributions informed by an analysis of 103.9 million GitHub pull requests. Its golden set combines human reviewers, frontier LLMs and static analysis, followed by independent senior-engineer validation; GitHub reports 96.6% agreement on golden true positives. The dataset, scoring methodology, judge configuration and self-serve runner are public.