As a new user I’m feeling frustration at having used 100/500 fast requests in a matter of the first 3 days of my subscription. I would like the option to choose slow requests during certain sessions when timing isn’t as important. This current model isn’t sustainable for me over the month. And no, I don’t want to buy more fast requests…money matters to me. It seems this would be an easy and popular feature to add. And, yes, I know there are other models to choose from but Sonnet 3.5 is the best at programming. I already utilize other LLMs such as Gemini 2.0 Flash, Claude’s chat, etc to help design, troubleshoot, and brainstorm before using the precious few fast requests in Cursor. Please add this or increase the number of fast requests to a reasonable number. Cursor is an excellent product but my subscription won’t be worth keeping at this rate.
Hey, while we don’t have this as a feature, you may find it better to switch to non-premium models, like DeepSeek v3, for the times where it fits well for the task.
Other models can be better than you expect as coding, but the temptation is high to always use the best model! I’d recommend experimentation to see how well other models work.
@danperks, My experience has been the same so glad to hear the feature is being worked on. Is there a projected due date for it’s release yet?
Not yet - I’m hopeful it will be soon, but I know that the wait will pay off once it’s ready!
“we don’t have this as a feature”
it used to be possible so you should say “we removed the option for you to choose”
Nope, we’ve never had the ability to choose between fast and slow requests in Cursor!
Slow requests have always been a backup to fast requests, only kicking in once your fast requests are fully consumed.
As I cannot prove it, I’m not so assertive about this but I do remember experimenting with this in order to control my rate of consuming requests.
Then they changed to Auto, Legacy and Fast
This was from a very old version of Cursor, and referred to the model that was working on a given prompt, not the speed of execution for the same model.
You can get the ‘fast’ behaviour by choosing a free model today that has no slow pool!
As we have no plans to add this in the future, I’ll close this thread for now, but I appreciate the suggestion!