I have used both extensively and I don’t care what the bench marks say - Claude has been way, way better and most importantly, predictable

I can handle issues much better if they are predictable even if the model makes mistakes — much more frustrating when the model is erratic

I find codex wanders off road more often and fails to see the “bigger picture” (as much as LLMs can see the bigger picture at least)

And tbh when it was first released Astral felt even worse

I’m being forced to use it right now and at the end of the day I’m making do so it’s fine, but Claude makes for a smoother experience