I think both of these points are what sow doubt in some people in the first place because both could be true if GPT-4 was just less profitable to run, not if it was worse in quality. Of course it is actually worse in quality than 4o by any reasonable metric... but I guess not everyone sees it that way.