I use and pay for both. Currently I use 4.6 (well as of yesterday) to do broad strokes creation. I use codex for audit. Generally first two or three audit cycles claude completes. There is often a subtlety that only codex can fix, but I usually do that at the end.

IME, codex is sort of somehow more .. literal? And I find it tangents off on building new stuff in a way that often misses the point. By comparison claude is more casual and still, years later, prone to just roughing stuff in with a note "skip for now", including entire subsystems.

I think a lot of this has to do with use cases, size of project, etc. I'd probably trust codex more to extend/enhance/refactor a segment of an existing high quality codebase than I would claude. But like I said for new projects, I spend less time being grumpy using claude as the round one.