GPT 5.4 xhigh thinking was really good at teasing out problems in multi step flows of a process I was refactoring, caught higher level/deeper problems than Opus 4.6. However getting it to write the code is just not a good experience for me, it changes the style/does not follow surrounding code, codes in a sloppy way and creates subtle bugs that I don't see from Opus. So I use codex for review and opus to write code. Testing the new Opus 4.7 still to see if the review/reasoning catches more/better stuff. I frequently fire off all 3 (Gemini 3.1 pro, Opus, Codex xhigh) on same code than have them cross reference each other and stuff like that. Gemini is so bad it's not even funny, not sure why I keep it running.