Cheetah: stealth model from an undisclosed provider

Discussions
25

Technically Auto could be using models that are not even in the official selection list, right?

26

pathetic and stubborn model response, doesn’t agree with you keep doing its own nonsense

27

@tester22 You can turn on privacy mode to ensure your data won’t be used for model training

28

This is critical information that is not clear at all from the ux, and I think is non intuitive, so this will help a lot - thanks. I don’t need military grade security, but I do need to know which model is being used, I had thought auto selected between enabled models, which I think is a reasonable assumption - after all, if a model is turned off, then it should not be used.

30

Hey guys, just to clarify a couple things:

  • Auto does not only use models you have enabled. It is built to withstand model downtime and provide you with a stable usage cost while giving you access to a range of the best models in Cursor
  • With Privacy Mode enabled, we have agreements with all model providers / inference partners to ensure your code is secure end to end. Once a query is completed, the model providers do not retain any of your code or conversation beyond caching (usually 1hr) and cleanup.
  • Models like DeepSeek are hosted with a US partner (Fireworks) and also abide by the data-retention agreement
  • Stealth models also retain this agreement by default.
7 Likes
31

Since Auto is charged at a constant rate, the user can’t influence which model it uses. Otherwise I could disable all the models except for sonnet 4.5 and have Auto only use that for much cheaper.

32

Got it - I think cursor should clarify this on the gui because it isn’t clear or obvious what auto actually means.

33

whilst I appreciate the response, having been insulted by another guy called Perky in this chat - whose messages disappeared at the same time your appeared, it makes me wonder if you are the same person.

34

then how were u trusting the openAI and claude?

35

I dont use openai. I only use claude. I have read the anthropic contract, I trust it, and it is the only LLM that I am allowed to use for some of my codebases. It is not just about security, I don’t want to test some random unnamed LLM, firstly, I am not allowed to, secondly because I want to use what I know, I don’t have time for beta testing LLM products. besides - I shouldn’t have to explain why I don’t want to. I don’t want to, and that is reason enough.

36

You can just choose Anthropic models and never use Auto. No one is forcing you to use the new model. And if it gets added to your list, it might accidentally be the selected model. But this usually only happens after you boot up, so just check it.