people have been complaining about this since GPT-4 and have never been able to provide any evidence (even though they have all their old conversations in their chat history). I think it’s simply new model shininess turning into raised expectations after some amount of time.