Cheating with model providing?

I see. But when you’re asking the exactly the same question on chatgpt[dot]com via the same “gpt-4o” model (which I use for API requests) you got the answer:

I’m based on the GPT-4 architecture

How to explain it? =/