I wanted to ask about whether you are using premium models with slow requests or not? Because i noticed a huge deference between 2 days ago and today. The agent is over the place and providing trash suggestions also, have no idea of what directory is working on.
Can someone from the support tell me what’s going on?
@danperks Please elaborate.
Hey,
Slow requests are always completed with premium models, the very same as fast requests! There should be no difference in the performance or output of fast vs slow requests.
If you use a smaller model, those are not subject to the fast/slow usage model, and should be completed fast every time!
Do you have some examples of the difference between the two?
Hey,
The only thing I can say is the fast requests are with clean code and nearly no mistakes but for the slow requests the response is dumber and full of bugs. so this made me wonder why (With fast requests my project runs smoothly but with slow requests, I fix bugs).
I’m pretty sure something is wrong After using the slow requests for a while. The responses are so bad. Also, I noticed the AI never looks at the rules in settings.
I would also like to hear a definitive Yes / No answer to this question. I am experiencing the same thing and I have the same question. Are fast requests using the actual assigned models, but slow requests are not? Yes or No? Cursor team should reply and give a definite answer to this question, please.
Did you figure it out?
Yes, it’s all premium after 2 months of use the models provide the same code and “dumb” suggestions. I call it dumb because we expect a lot from AI and think it will build us what is in our brains without fully explaining the user flow/features/function/architecture.
My advice after I learned this the hard way 
Just split your functions into small pieces so the AI understands your requirements AND GIT, GIT, GIT, GIT