My experience with Pi was that I pasted some error and asked Codex to solve it and it explicitly said something like "you could be mistaken into thinking this is X but it's actually Y", which was weird because it was pretty clearly not X and I've never seen it say that. Then I installed Pi with the same OpenAI model and gave it the same task and it told me it was X and I promptly never used Pi again because why would you want to take chances like that.