Claude is better for longer code IMO

ChatGPT is more consistent for me

Depends on the task tbh

In my experience codex has been much better at understanding huge code bases and finding bugs. But I haven’t used CC for a few months now so it might have improved. Their models got so degraded for whatever reason (either quantization or just their midlayers)that I came to codex and I have been happy with it. Codex can take longer but I can trust in what it says more than CC, which would confidently tell me something is fixed and not actually do anything, or filling up my entire code base with random MD files I never asked it to make. You will also use more of your limit if you keep getting bad output which happened a lot more in my experience with CC.