My Premium models = 500/500
gpt-4o-mini or cursor-small = 330 / No Limit
I understand I am out of “fast” requests. Cursor still shows its selecting Sonnet 3.5 however I came out to my account, I see my mini going up.
I am quite happy to wait for a “slow” premium request and do not want to use the fast mini as I’m trying to code and it uses poor logic.
How do I set it to favor “slow” premium requests?
I do not see this section under my account. Where do I find this view?
oh I see, I have “enable usage based pricing” disabled.
Do I need to have it enabled to get “slow” premium?
What we see is that the Anthropic/Sonnet API is not being called at all in agent mode, and that cursor-small is being called even though we should have unlimited free slow premium model requests.
No, you don’t need to enable this option to use slow requests. You need to switch it to view recent usage and determine which model you are using at the moment.
I am also a non-pro user and would like to know how to continue using cursor with slow requests. I do not wish to pay right now until i have clear understanding if app development is my thing at all.
So how do I get it to use “slow” requests? Instead of fast dumber requests?
This is slow.
On the free plan, there are no slow requests, you are only limited by the number of requests allocated to you per month.
I am a paid user. I have used my 500 and now am getting over 300 mini. Even though I have sonnet selected. I do not want to use the mini, I would prefer to wait for slow premium. How do I do that? The logic of the mini goes in circles.
Any update on this? Any Cursor staff here? This seems like a major bug and is basically false advertising. It is currently impossible to trigger a “slow premium” request using the “composer agent”.
In the same boat, payed as a Yearly user because of that Advertising - now it is a bit unlucky that this is going under the radar. The “unlimited slow requests” was actually why I decided for Cursor and not another competitor, now they just take it away without saying anything is disapointing 
There is an other thread about this saying it is because they cant sustain the cost for it, still hoping for some official Information on this topic.
Hi, it’s a bug actually, i can relate that i use slow requests for 3.5 Sonnet since 2 days now.
But actually the slow mode is really slow since yesterday…
Hey all, as long as you have Claude 3.5 selected as your model, that is the model that will be used, regardless of whether you have fast or slow requests available.
While the LLM may say it’s an AI coding assistant designed by Cursor (as shown in @maxcurrent420’s screenshot), this is actually Claude responding! We do tell it that it is a coding assistant designed by Cursor, but it seems to have got overzealous in its statement that it operates exclusively for us!
As @solariz linked, we are having issues with Anthropic not providing us with the capacity we need to sustain Claude requests like we’d hope to, which is causing the slow queue for Claude 3.5 Sonnet to be longer than we’d like, but your requests will still be answered by Claude when you choose it.
Alternatively, using a different model, like gpt-4o, should have a much smaller queue!
I am a paid user who has hit his 500 limit. I am only using sonnet 3.5. I have sonnet selected, but it is running gpt-4o-mini. I see the requests ticking up. I have used 330 fast requests when I want to wait for sonnet slow premium instead of fast. Are you saying I should have zero fast requests if I’ve only selected and executed against sonnet 3.5?
Hey, we use GPT-4o-mini for some behind-the-scenes things, which likely explains the usage increase. Also, there can be a delay between when you do a request, and how quickly it shows in your dashboard.
If you have used all of your 500 fast requests, all of your requests to Claude will be slow, but will still be completed by Claude 3.5 Sonnet, as long as that’s the model you have chosen.
@danperks Thanks for the follow up here - would it be possible in the UI show a short info under the reply which model acutally was used? I still very much the opinion that sometimes the “SLOW Request” with Sonnet still fallback to something else.
Just today I had several Sonnet Agent answers which were clearly way under sonnets capabilities and genereted garbage code which were not actually even runable. Things I normally only see from small mistral or dolphin models I play around with. And it wasnt a complicated task.
If it was sonnet then for sure the quality difference is alarming.
Another thing, I normaly very happy with my credits on fast queries per month, but sometimes I would buy more. Would because currently they are not taken over in the next month. So if you buy another 500 Credits on 20 Jan it is actually a waste. Is there any plan to let me carry over extra bought credits in future? I mean not forever but at least have them 30 days after I buy them.
Hey, if you had Sonnet selected in the UI, then the response was generated by Sonnet. We have no functionality to redirect prompts to a different model than the one you selected.
While have no plan to allow you to carry over requests, you can now enable usage pricing for premium requests, where you are charged $0.04/request to a premium model, like Claude 3.5 Sonnet. Therefore, instead of buying requests in batches, you can simply enable usage pricing and use requests as you like. At the end of the month, you will be charged only for the requests used outside of your default allowance.
thanks for clarification. The pay by use option is a good alternative.
