GPT-5 bombed my coding tests, but redeemed itself with code analysis

▤ Summary

This article compares GPT-5, GPT-5 Pro, GPT-5 Thinking, and o3 on code repository analysis tasks. It shows that GPT-5 Pro gives the most actionable, implementation-level insights, while base GPT-5 still delivers strong breadth and useful security and architecture observations. It also argues that better prompting can push base GPT-5 closer to Pro-level output. The piece concludes that GPT-5 is a meaningful improvement for code understanding, even though it fell short in coding tests. It frames AI code analysis as useful for onboarding, refactoring, and security review.

▦ Linked Companies

OpenAI

$25M to $50M