The best ai code reviewer catches one in four known defects. github's reviewbench, out oct 5 on 219…
the best ai code reviewer catches one in four known defects. github's reviewbench, out oct 5 on 219 real prs, gives copilot 87.8% precision but 26% recall, and codex 81 critical catches to copilot's 63.
review is the new frontier and nobody is close to thorough.
Context
GitHub's blog post of October 5, 2026 launches ReviewBench, an open benchmark for AI code review built on representative pull requests, a multi-source golden set and production-aligned metrics. The BenchLM leaderboard page says it tests complete code review agents on 219 pull requests, and AICoder's summary says they come from 187 repositories in 19 languages.
Per AICoder's summary of the leaderboard, Copilot Code Review in its Balanced configuration has grounded recall of 26.0 percent with 87.8 percent precision, ahead of Devin at 23.8 percent, Qodo at 22.1 percent and Cursor at 9.6 percent. BenchLM notes that GitHub publishes the benchmark and evaluates its own product on it, that Claude Sonnet 5 acts as the judge, and that the reference findings can be incomplete.
The 219 pull requests, 87.8 percent precision and 26 percent recall match the sources read. 'Catches one in four known defects' is a fair reading of 26 percent grounded recall, with the caveat from BenchLM that the reference set can be incomplete, so recall is measured against known findings, not all defects.
'The best AI code reviewer' needs care. GitHub built the benchmark and its own product ranks first on it. The New Stack reports that an independent leaderboard run by Martian ranks the same product differently, and its headline says the independent one 'tells a different story'.
The note's 81 critical catches for Codex against 63 for Copilot were not found in the pages read, so they are unsupported here, not refuted. AICoder lists a Codex configuration at 19.9 percent grounded recall.
'Review is the new frontier and nobody is close to thorough' is the author's line.
Related work
- ReviewBench: An open benchmark for AI code review (The GitHub Blog, October 5, 2026) ↗Source for the benchmark's design and GitHub's use of it on Copilot code review.
- Copilot tops GitHub's own AI code review benchmark. An independent one tells a different story. (The New Stack, October 6, 2026) ↗Independent reporting that questions the first-place result.
- ReviewBench Leaderboard and Scores (BenchLM, October 2026) ↗Leaderboard page with the 219 pull requests and the conflict of interest note.
- GitHub open-sources ReviewBench: Copilot, Devin and Qodo lead (AICoder, October 6, 2026) ↗Summary of the leaderboard recall and precision figures.
Watch next
- Find the per-severity breakdown behind the Codex and Copilot critical catch counts. Compare ReviewBench with the independent Martian leaderboard.
Sources
- ReviewBench: An open benchmark for AI code review (The GitHub Blog, October 5, 2026)github.blog
- Copilot tops GitHub's own AI code review benchmark. An independent one tells a different story. (The New Stack, October 6, 2026)thenewstack.io
- ReviewBench Leaderboard and Scores (BenchLM, October 2026)benchlm.ai
- GitHub open-sources ReviewBench: Copilot, Devin and Qodo lead (AICoder, October 6, 2026)aicoder.com
Provenance
The note above is reproduced unedited from the original post, first published on Threads on 9 October 2026 at 09:37 IST. Sources are the papers and datasets the note draws on.
View the original post ↗Embed this note
More notes
The air is now being asked to keep its own ledger
the air is now being asked to keep its own ledger: ecmwf’s aifs compo becomes the first ai model to forecast atmospheric composition globally every three hours, cleanair simulates 365 days of pm2.5 over china in ten seconds, and a unified framework maps six pollutants at one kilometer across the whole country. the air now files its own composition report.
read the note →The current is now being asked to draw its own map
the current is now being asked to draw its own map: china’s langya 2.0 predicts six ocean phenomena including internal waves and mesoscale eddies, a deep net called wenhai resolves eddies globally with air sea flux formulas built in, and scripps infers surface currents from the way temperature patterns deform in satellite images. the ocean now files its own circulation report.
read the note →The soil is now being asked to report its own carbon
the soil is now being asked to report its own carbon: a nix color sensor paired with generative data augmentation predicts soil organic carbon without a lab, random forest drives 74 percent of soil health mapping studies, and sentinel 2 tracks five year carbon change across france and italy from 922 samples. the dirt now files its own carbon account.
read the note →