I chose to use the Gemini 2.5 model, but Cursor is using Sonnet 3.5 (the model I used in the previous conversation) to respond.
Also, it’s very inconsistent. Sometimes it’s 3.5, sometimes it’s Gemini. I don’t know if it’s random. I tried saying “hi” and retrying continuously and noticed that.
Same for me!! No matter what model I choose it says it’s Sonnet 3.5 and also keeps repeating itself when generating a new message. Says the same thing 1:1
This issue is super serious. What’s Cursor up to? Are they using us for blackbox testing to see which model is actually good, or is this a scam to boost profits?
I’m also being directed to Sonnet 3.5 with the claude-3.7-sonnet-thinking-max model I’m using, which has wasted my time and money.
Just tried Deepseek, and it randomly picks between 3.5 and 3.7 for responses =))
Please share some request IDs so we can check on these queries. If you’re in privacy mode, would be great if you can temporarily disable, perform a query, and send us the request id
Hey all,
I cannot reproduce the model saying it is Claude if I have Gemini selected! You can always sanity-check which model you are talking to by seeing it “thinks” and specific to Gemini, see if it thinks after every action (think → reply → edit file → think again, etc).
We are currently seeing some weird issues with Gemini 2.5 but we don’t currently see an issue on our end. See this thread for ongoing discussions:
Hey, here’s how to reproduce that bug:
Open Cursor in one project, select the sonnet 3.5 chat model, and cancel repeatedly.
Then, open Cursor in another project (so you have 2 Cursor tabs running at the same time) and select deepseek v3.
The response will be sonnet 3.5 instead of the deepseek you chose.
Any updates? I’ve been exclusively using Sonnet 3.5 today, regardless of the model I choose, including deepseek-v3, it still uses Sonnet to respond.
