I recently upgraded to a Pro subscription, and while I’m working on mapping out my project details, structure, or roadmap, I wanted to switch to slow request mode instead of using up my quota in fast request mode.
This way, when I move on to coding tasks, I can switch back to fast request mode and work more efficiently and effectively.
However, I couldn’t find such a setting, and unfortunately, my quota is being reduced while I’m having less important conversations. When I reach the most critical stages of coding, I’ll be forced to rely on slow request mode because my quota will have run out.
It would be great if users could manually change or select this mode as needed.
Or, if I’m misunderstanding something, please feel free to correct me.
Thank you for considering this, and I truly appreciate any steps toward improving the usability of this feature for Pro users
I noticed this as well. I upgraded to Pro and was trying to work out why the Chat and Composer couldn’t access my files and I wasted about 20 premium requests asking things like “can you see my @ blah.md file now?”.
I realized after that I could switch the model for basic things to use the o1-mini or w/e but I honestly don’t know the effect that will have if I do some basic setup with one model and then switch to Claude for the code changes.
I totally agree, I have the same use case, and the only solution for now is to connect to the Cursor website, into your account, and enable or disable the fast requests.
In the same way, there is also a missing switch to easily use custom models or your own OpenAI key.
For example, in your case, you want to use the Fast Request mode to have Cursor handling your credits to call Anthropic, but you can also have the way to input your Anthropic API key into the model and enable or disable this directly from Cursor. And for me, it’s less painful because it’s just inside Cursor. No need to do it inside a browser but it’s still painfull to switch.
The best solution, at least for me, could be to have the list of available models in the chat or the composer and instead of just
“Claude-3.5-sonnet”
the list could be:
“Claud-3.5-sonnet with your API key”
“Claud-3.5-sonnet with fast requests”
“Claud-3.5-sonnet with slow requests”
This way, no need to enable or disable an API key and model names, just selecting the right model and use case.
I am stoked for the cursor team to open the gates for us using a clone of the deepseek models. Then we can all run fast and play without chains Isn’t that right @deanrie !!!
This is definitely something that needs to be fixed. I should be able to decide how I use my quota. I already have to use it up within a month, so at least let me decide the timing.
I hope they fix this quickly. Honestly, it’s not a huge issue and could be resolved pretty fast—if they want to fix it!
Also, when you close and reopen the program, past conversation history gets wiped, and you start with a clean composer page.
This is a much bigger flaw and needs to be fixed immediately.
Yeah, I came here to see if that was possible because I am going to use up my 500 requests in a few days, more or less. I have to go back and forth fixing a lot of stupid errors and it’s a big project. I’m often doing a bunch of things at once, though, and it will be several minutes before I even look back to see what Cursor did. I could certainly wait longer a lot of the time and would love to opt to do that.
Also, there is another point that frustrates me, and I feel like the Cursor team is doing this intentionally.
Whenever it makes a change in a code file, despite my repeated warnings (saying “share the entire code, not just the modified part”), it still shares only the code block and responds with something like, “I can share the full code if you want.” This results in wasted credits every time. Out of 180 credits, around 50 might have been spent just to get the full code on a second request.
If you say this technique is used to optimize for less workload, then you also need to explain why it forces an additional request that wastes 1 extra credit every time. If I spend 1 credit anyway, wouldn’t that mean there’s no need for optimization?
I think you shouldn’t focus so much on maximizing profits if you want users to keep their subscriptions. If you keep doing this, I will most likely not renew my subscription for a second month.
I kindly ask you to set AI free, release it from these shackles.
I’ve wasted most of my requests already just going back and forth on stupid things that I’ve already tried to correct or add to my .cursorrules file. It was cool at first, but it’s going to cost me hundreds of dollars a month at this point to keep using it.