Auto performance is very bad

Support Bug Reports
, ,
1

I think indexing is broken or something because auto is performing SOO bad.

I told cursor to implement something and instead of using an existing client interface, it was writing out all of the implementation inline.

It feels like it has lost context or is not using its context or something. Its VERY noticeable. Almost every single request, its doing something crazy, not like it used to.

Just a few days ago, I would give it little direction and it was one shotting everything with great performance. This feels like an index issue. Its not going off of existing patterns.

1 Like
3

I mean did something happen? Its horrendous! It like has like 0 context of the code base. It cannot solve anything. I gave it a task that was like 10 or 15 line change. Auto has not performed like this in a very long time.

4

You should include your request id, cursor version info and stuff to get your issue looked into.

By the way, why use auto? It doesn’t guarantee a good model. Use specific models instead - if you want to know which models might suit your usecase take a look at https://artificialanalysis.ai/ and browse through the various indexes. Different models work best for different usecases.

However, I’d warn you to stay away from Gemini models as they have the highest hallucination rates. Stick to GPT-5.2 or Sonnet/Opus for writing code.

5

Hey, I’ll need a bit more info to troubleshoot:

  1. Check the indexing status: Cursor Settings > Indexing & Docs. What’s the current indexing status?

  2. Get the Request ID for one of the bad requests: in the chat menu (three dots in the top right) > Copy Request ID.

  3. Cursor version: Help > About Cursor, or Cursor > About Cursor (Mac).

  4. Rough size of the codebase (number of files or lines), and what the main language is.

Also try:

  • CMD/CTRL + Shift + P > “Developer: Reload Window” to reload the window
  • If indexing shows issues, try re-indexing: delete the .cursor folder in the project root and restart Cursor

About Auto mode: it picks the model automatically, which can lead to unpredictable results. If this is critical for production work, you may want to switch to a specific model (Sonnet, Opus, or GPT) directly via the model selector.