Opus 5.5 routinely finds tons of bugs in code that Opus 5.5 wrote, so this may not reflect a difference in model capability.