Skip to main content

Blog

Articles & Notes

Notes and experiments on software engineering, AI, and developer tools.

·3 min read

GPT models for code review

I tested seven GPT models as code reviewers on four pull requests, twice each. GPT-6.1 Sol caught the most real bugs with the fewest false alarms, costing about $0.24 per PR.

GPT-6.1 Sol: 4 bugs/round~$0.24 / PR
Privacy & analytics

When enabled, Vercel and PostHog measure page views and project interactions. No session replay or visitor profiles.