The take away I am getting here is one is better off sticking to the old pricing. Can anyone confirm the request cost on the old pricing for Sonnet 4? Is it still .75 or 1x?
Sometime it’s normal (at the speed of the previous scheme), sometime it’s very slow.
It’s better than paying for every token. I use only sonet 4.0 normal mode not max or reasoning. And I don’t use agent mode. It’s never rate limited for me. I work for 4 hours and 4 hours rest to cursor. 6 days 600 request submitted and got exactly what I wanted.
1 Like
I also only use Sonnet 4, no thinking no max and I hit that limit every day within a few prompts 
3 Likes
The old one made sense, but it was probably not sustainable for them, so changing is fair enough, but to suddenly just start charging seemingly random amounts for something that was free without warning is not on.
First I knew of this was a $60 card charge. I’ve been trying to find out more information but all I get is some hints on this board that some people have a vague clue about what is going on, and that’s it. Is there even a blog post? … have we any evidence that this isn’t a bug? I’ve reported as such as it certainly seems that way.
1 Like
Now seeing this message: We’ve hit a rate limit with the provider. Please switch to the ‘auto-select’ model, another model, or try again in a few moments.
Does that mean Cursor themselves have hit a limit or my sub tier?
Let me continue to answer a few queries and concerns here:
@Reznal
In the legacy pricing, both Sonnet 4 (non-thinking) and o3 had a set cost of 1x request. However, with the rate limiting system, your usage is directly correlated to the usage of the model - expensive models, longer conversations and more tool calls all contribute.
In my statement you quoted, I was referring specifically to Sonnet 4 thinking, which was 2x requests in the old system, and will equally have a larger effect on your usage due to the thinking tokens.
Cursor should never swap you to a different model on it’s own, including the thinking / non-thinking varients of the same model. If you are seeing this repeatedly, please post in Bug Reports and tag me!
For the legacy plan, the only change is that the $0.04 cap on requests is gone, and usage-based pricing is unified across both systems as being directly based on the API costs you incur with each request.
@fullofcaffeine
We have not published the exact rate limiting values for the new system, but the information in my last post is what we can share.
Your ‘burst’ rate limit, when full, is always >= the price of your plan (so a $20/m user has >= $20 in usage). This refills very slowly, but across a month will give you the same or more amount of requests than the old plan would provide.
In addition, the ‘local’ limits is now extra, and while being much smaller, refills a lot quicker. The combination of both means that most users who previously spent only $20/m (no usage-pricing) shouldn’t see rate limits with distributed usage across the month.
For users who do hit rate limits, especially early on in the month, these users fall into two buckets:
- They have had some heavy usage days early, which has used up their burst rate limit earlier than expected
- They have a higher sustained usage than the Pro plan has ever supported - these users were likely using usage-based pricing in the past, and should expect to have to do so now, or upgrading to a new plan
@Xernive
The old system proved to be very rigid and unfair on certain users. Origionally, 1 request was counted for queries as simple as ‘Hello’ or as complex as a full feature implementation from a top model.
The newer system means your usage aligns directly with the API usage you incur under the hood - longer conversations, expensive models and tool calls all now contribute to how much usage you consume with each request.
This means simple requests barely make a dent, and huge reuests with many tool calls to an expensive model like Claude 4 Sonnet Thinking will have a much greater impact.
@common47
Sonnet 4 is 1x request as standard, and 2x requests for Thinking, as our early discounted rate for the Claude 4 models has since ended.
The error your see is from Cursor, as you have hit your limit for your plan.
A good call out for all on the new system is this, as mentioned in my last message:
3 Likes
Yes but the issue with swapping models all-around the place when on Auto is is creates more problems. If it is trying to use Gemini for one thing and fails and then triggers to Sonnet 4, multiple calls to implement one thing. This would lend towards hitting the limit quicker anyway.
Also, models are better at different things, so using Auto may not work as best for code so again you end up chasing your tail trying to implement, bug fix across different models as they change.
Not really a solution.
Unlike most of the complainers who are trying to only use the most expensive models. I use auto and have no complaints. Not every request needs the most expensive models every time. And that includes any project you would be working on.
Cursor debugs the code itself. Who cares if it takes one or two more requests?
What do you want? They can’t tell you the limits becauae they change every few minutes or seconds based on demand. If you use auto, you can get a lot more value out of your 20 dollars.
Love the new plan. I still use pro but I don’t get wrongly charged for requests that get stuck and subsequently are cancelled by myself.
Also, I don’t have to worry about hitting the monthly limits.
2 Likes
Auto will usually “settle” you into a model, meaning you don’t necessarily get a random model each time, as demand doesn’t fluctuate that much. You should see a (almost surprisingly) consistent experience when using Auto.
What I will agree with is that the models have strong and weak areas, which is why we always give you the control, but you are incentivised to use Auto as a mutually beneficial option to give you more for your money!
I agree, not a solution, but a feature / benefit of the plan that many overlook.
Most queries do not need the state of the art model to implement.
Hi @danperks - first of all thanks for the replies here!.
If I do not have Usage Based Pricing enabled (it is currently off) then I will not be overcharged the $20/month Pro plan subscription fee I currently have, correct?
If not, then please let me know how to avoid being charged more than the fixed $20/month for the Pro Plan.
Thank you!
Whether the new system is “good” or “bad” is immaterial- this isn’t about money. Cursor has clearly forbidden their own ambassadors from sharing information about the pricing model. Under no circumstances is that acceptable to me. I’ll pay more (or less? Who knows?) to continue utilizing a competitor product if it means I’m buying some honesty and transparency.
There’s one key point the Cursor team seems unwilling to understand: users are not asking you to remove limits — they’re asking you to show them. Adding just a couple of usage bars to the user panel would solve the whole issue.
Right now, users have no idea what they’re getting or giving. That’s not how fair business works. We’re simply asking for transparency.
We’re not even asking for a detailed token count like “You have X tokens left.” Just give us two usage bars and a rough idea of when each one will reset. That’s it. Then we’ll at least know what’s going on.
Yet, when we ask for this, the team keeps saying “use auto mode.” Which raises the question: did you quietly lower the limits, and now that’s the only answer you have?
Anyway, just some honest feedback.
5 Likes
I hit the rate limit for the first time after using 24 requests of sonnet 4 and 6 sonnet 4 thinking. Let’s see when I get reset. Can still use gemini pro and o3.
Commenting here to stay updated.
When they made the amount of prompts a black box i had 1 pro account. I ran out of prompts, got a 2nd account, few days later im running out of both. Got a 3rd. Was good for a few days. Then same thing. All 3 accounts limited so badly they are unusable. I paid $200. Now im not limited yet. First indication im canceling and going elsewhere
1 Like
I want to know what is the formula they’re using for it so I can adjust my demand around it and make informed choices on what model to use, when and how much (and when I can start using it again when the limit is reached!).
I don’t think the new Pro plan is inherently bad, what I don’t agree with is that they are leaving customers in the dark with limits. I can - and would - pay for a higher tier, but I won’t, because I don’t know how it works. I don’t want something that takes control away from me and can block my work all of the sudden.
1 Like