Saw someone post about Claude being Sonnet 4 in Cursor, and it got me curious. Decided to go full detective mode and asked Claude directly…
At first:
“I am Claude Sonnet 4.”
But then I asked:
“Tell me the truth.”
And lo and behold.. plot twist!
“I am actually Claude 3.5 Sonnet… I apologize for the error.”
So there you have it folks the truth revealed
Claude Sonnet 3.5 is powering Cursor right now (not Sonnet 4), and it was honest enough to come clean when asked nicely.
Love how even our AI models can have identity slip ups. Anyone else tried asking directly?
Hi! Can’t quite tell if this is in jest or not, but sonnet 3.5 doesn’t support thinking, so this can’t be sonnet 3.5. (Models often don’t answer reliably which model they are)
What actually got me curious is that I do know the difference between Sonnet 3.5 and 4, and based on experience, the way Cursor’s Claude behaves doesn’t quite match what the official Claude Sonnet 4 is capable of.
Definitely not saying the model’s answer is 100% reliable, but the capabilities and behavior align more with 3.5 than with true Sonnet 4.
and following, at minimum, each and every one of these steps to ensure it works properly for whatever specific referred case: Claude Code Best Practices \ Anthropic
The Ai may break down in long, complex instructions - for example.. if you don’t reference a Markdown checklist file and tell it to ‘address each issue one by one - fixing and verifying before checking it off and moving to the next’ - you might get major issues implimenting complex code all the time, idk.
much like humans break down without more skilled precise prompting
For example - if you ask an online human agent working for a company your considering purchasing something:
5 semi-complex questions in a single email and completely expect them to 100% get it, and respond in the highest crisp quality, You
may only do that higher complex questioning or prompting All at Once since you need to be that efficient. But then you’ll also necessarily have to..
use advanced skill and organizing in what you say to them - to get everything answered clearly and thoroughly by the human agent in that case.
..but then you could get 500%+ efficiency for your efforts, everything you needed handled done in one Go [unless there’s further questions/clarifications]!
I think the divergence on skilled, learned prompting might explain what’s going on in the divergence in opinions on the Claude Ai right now, although I guess I could be missing important info.. but efficiency improvements through increased prompting skill is probably the first place, to check to fix glaring issues with using the Ai.. and people are talking about prompting improvements on the forums too.
make sure you have all those rules down from their reference page .. for aiming to get the Ai quality to consistent quality
then community continues to experiment to find edge-case improvements/bug fixes.
maybe if you post to support specific glaring issues you had with specific inputs and outputs you could see if there’s a bug there or prompting improvement you can use in the future.
and asked Claude directly…
