The model clearly is not sonnet 4. I noticed the quality of the replies were not as good as they should be and guess what, someone has made a swap. I think if that is intentional we have a BIG problem because the model says it is not sonnet 4 it is sonnet 3.5.
I went up to sonnet on the desktop and asked the same question and it knew it was sonnet 4.
Steps to Reproduce
Just ask the model what it is.
Expected Behavior
I expect when I pay premium prices I get premium service!
When you select a specific model, the request is always processed by that model without switching. So if you choose Claude 4 Sonnet, the response will come from that exact model.
Important: all Anthropic models are named “Claude”, and Claude 4 was trained on data that included information about Claude 3.5 Sonnet. Also, models accessed via API without a system prompt have no knowledge of themselves.
I’ve tested this multiple times, and the results are consistent.
ya know we can make an llm say whatever we want? CC might just have it in a sys prompt, ya know? why, you ask? so you people stop complaining that a mere token prediction model isn’t self-aware or even try to make conspiracies that the provider fakes a model.
they can make sonnet say it’s ■■■■GPT-9000-pro-ultra ffs, but why should they even put these useless tokens in the prompt if it DOESN’T improve actual model use?
When asking these self identifying questions to models in cursor, they used to respond with something along the lines: “I am not allowed to share any information about the source model” I guess they got rid of it because it seemed obviously shady like they were hiding something. So now we get this response that says its a different model than what it is, but from your test sonnet-4 can self identify. Even if Cursor is giving us different or some composite model, why would they not just tell that model to at least replicate the self identifying behavior of the actually requested model. Why don’t they just make it say its Sonnet-4, even if they are using a different model or some composite one?
The chat history does play a role in what the model identifies as. Like you can start off with gpt-5, and it identifies “correctly”, then change to auto and it will keep saying that, but then change to Sonnet-4 and it will now say it is sonnet 4, all in the same chat. So Auto possibly hides who it is and just says what the previous chat was, or they give you the same model in Auto, which is doubtful.
Now mine is just getting scrambled or Cursor, for Auto at least, is genuinely creating some kind of composite model. Claude-4o-mini
