Claude Opus 5 fixes 11 of 45 hidden bugs, versus 2 for Opus 4.8

PawelHuryn · x · 2026-07-25

A user reports that Claude Opus 5 fixed 11 of 45 bugs hidden in their own 28K-line VS Code extension, while Opus 4.8 fixed 2 under the exact same repo, prompt, and setup.

They say Opus 5 not only fixed more bugs, but also found the same two bugs as 4.8 plus nine harder ones, with zero false fixes. The model behaved oddly in places—ignoring “analyze this solution” and inspecting git history—but that detour led to useful insights. The poster calls Bug Hunt Bench just one test, but says Opus 5 is now their favorite model.

Related event: Bug Hunt Benchmark: GPT-5.6 and Opus 5 Lead in Code Repair(2 posts)→

Original post →

More from coding & agent

coding & agent channel →