In my opinion, Cursor is the greatest and most powerful tool for building and coding. Recently, in these past 3-4 days, the quality of the AUTO Model has dropped drastically; its ability to reason, understand context, etc., has made building almost impossible for me now. Let’s not even mention the time it takes to receive a response. I’m waiting up to 1 HOUR for the agent to finish a task, yes, my chat is long, but it did not have a single problem whatsoever 3-4 days ago. So I HOPE we can see Cursor getting back to its usual greatness as soon as possible, but something certainly has to be fixed. Have you had similar problems recently?
Hi @TarkanErtan I’m sorry for the frustration. We are actively working to improve Auto, and the changes you’ve noticed this week are likely directly related to that. There have been several underlying changes that may affect how auto feels as we refine and improve it. I’m sorry this experience has come up short for you.
I do have a couple things that can help tho. I noticed that one of your chats was 100 + minutes long, which can amplify the negative effects you’re seeing.
If you want consistent, predictable behavior, pick a specific model from the model selector instead of Auto. That pins every request to the same model, so you’re not affected by Auto’s routing changes.
For the slow responses, conversation length is the biggest factor. Every request has to re-read the entire chat, so a very long thread gets slower (and can start to lose focus) as it grows. Starting a fresh chat for each new task, rather than continuing one long conversation, should cut the wait time significantly.
Thanks a lot, that fixed it indeed. Opening a new chat significantly improved waiting times, and switching to the same model (Sonnet 4.6) resolved the quality issues. Although I’m now spending more on API, previously, when I used AUTO, it would deliver a similar quality without using up the API percentage. Looking forward to seeing the improvements to AUTO mode. thanks!