Today:
GPT 5.5 - is a perfectionist that never finishβs job
Opus 4.7 - is an average full stack that finish everything with a lot of bugs and security concerns
Honestly,
The saddest thing about Codex is that it never has the ability to finish a project (and I spent a few billion tokens per month):
1) Either it only does one step at a time and we have to endlessly type "continue" - and even then it never finishes.
2) Or we use /goal and it gets lost along the way and wastes a lot of tokens without finishing.
3) Or we have to do the "heavy lifting" with Codex and pay good devs to polish the front/back/infra, etc.