That’s falsifiable quite easily by measuring tokens per second. Rather, the real reason codex takes longer is that it does more work to read more context. IMO

That’s falsifiable quite easily by measuring tokens per second. Rather, the real reason codex takes longer is that it does more work to read more context. IMO the results are much better with codex, not even close