I don't care about "personality" I want quality.

Honestly when I get codex to review the work that Claude does (my own or my coworker's) it consistently finds terrible terrible bugs, usually missing error handling / negative conditions, or full on race conditions in critical paths.

I don't trust code written by Claude in a production environment.

All AI code needs review by human, and often by other AIs, but Opus 4.6 is the worst. It's way too "yeet"

The opus models are for building prototypes, not production software.

GPT 5.4 in codex is also way more efficient with tokens or budget. I can get a lot more done with it.

I don't like giving money to sama, but I hate bugs even more.