Indeed there is still confusion because I’m sorry what, I still don’t get it @condor. Jrista is saying Pro, Pro+, Ultra are still in general using requests (regardless of your usage or Auto mode), but they are not, right? Nowhere here Cursor – Models & Pricing or here Cursor – Models does it mention those plans are using a request based model, but that they’re using token based model at API rates. Only teams are currently still using request based model and even that will change to token based model at API rates come 15 september. Or am I just stupid and completely not getting the terminology here?
@plesknekekec what Jrista mentioned are Legacy Pro accounts that count requests. Current Pro, Pro+ & Ultra are not counting requests but tokens as reported by AI providers. That doesn’t change, we listed here on top of this thread for those legacy accounts their cost as rest is by api.
Yes teams are using request counting until 15th Asept.
Aha. And legacy pro accounts are those who had annual plans before that change from requests to tokens was made, whenever that was?
Legacy Pro are those that requested to remain on request based counting during the June/July period, that wont change.
I’m subbed since June, so I could’ve requested (or still can?) to remain on request based counting for as long as I remain subbed?
Ah, ok. I was mistaken that the current plans were request based.
So I think I understand it now. That said, i think it is still a rather confusing issue. I am not sure where it would be most appropriate, but I feel a single, reliable, consistent location where the pricing (now and after reported dates when pricing will change) is explained, for all plans, including any grandfathered in factors, is explained all at once, and in clear and concise terms.
IMO I think there is some terrible confusion about how Cursor is charging, and I think that may be leading people away from the product thinking its way too expensive, when it is not as bad as it seems. I can’t speak for every country, but I think for most professional software devs and companies wishing to incorporate agentic coding into their workflow, Cursor is…reasonably priced. I don’t think ANYTHING with AI right now is actually priced at a truly affordable cost, and the cost seems to be rising, but it is a new and highly disruptive industry, and I understand it takes time for things to settle.
Given the rather chaotic and confusing nature of the industry, a clear and concise explanation of Cursor costs, I think, would help settle your customers (both current and potential) fears or concerns about cost, now and longer term. I myself, have concerns, and I am not entirely confident in my own understanding of the costs of using your platform (I’m on a plan, but once that plan usage runs out, I am not entirely sure what kinds of costs I’ll be facing, TBPH.)
Anyway, hopefully you guys can sort it out and give your customers a clear picture! 
3 Likes
Yup I very much agree, a comprehensive post would be very useful, especially for noobs like myself.
Even stuff like explaining that you probably shouldn’t be using Claude 4, yet alone thinking models, for the simplest defined prompts or dumbest little requests, and should instead use cheaper models and learn to understand their pros and cons for various tasks. Just how for example there is no need to use GPT-5 for a majority of simpler tasks and you can instead use GPT-5-Mini, define the task well and get the same results but for 5x cheaper price. And for the most simplest tasks or high-throughput tasks, especially simple instruction-following or classification, you can use GPT-5-Nano at 25x cheaper price. This is all even explained on OpenAI website, not to mention that prompting is a skill, but ofc most noobs and vibe coders do not bother with that kind of research, which ofc you could argue is their own fault and part of Darwinism, but if everyone keeps complaining in the forums it does hurt the brand somewhat.
1 Like
Aye! The model selection task is really important. This is an area where Auto could be improved as well (especially if we are going to be paying for its usage, in whole or in part, in the future). Right now, you have no control over what model Auto will use. I think I opened a feature request for this here on these forums, but, it would be nice to be able to provide hints to Cursor, around which models we PREFER to be used for a given kind of task (and maybe even ranked):
- Planning: GPT-5 (thinking), Gemini 2.5 (thinking)
- Coding: Claude 4 Sonnet, Claude 4 Sonnet (thinking), Claude 4 Opus
- Basic Research: GPT-4o, GPT-5 Mini, GPT-5 Low, DeepSeek
- Advanced Research: GPT-5 High, Claude 4 Opus
Something like this would at least help “Auto” choose a model better suited to the task, and the COST EXPECTATIONS. Cost has now become a FUNDAMENTAL aspect of choosing whether to use an agentic IDE or not. It cannot be ignored.
Further, if Cursor takes the path of pushing everyone towards the most expensive models all the time, users ARE going to notice that, and they WILL act on it (i.e. by jumping ship, or reducing in-cursor agent/model usage in favor of external model usage, say ChatGPT or Claude web apps).
I don’t know how any of this helps Cursor, Anthropic, etc. become profitable, but, if they don’t address it somehow, the cost factor, I fear they will lose more users than they attract in the long run.
1 Like
I absolutely want to get back on request based pricing quoted in the original post.
It’s unfair that some can still benefit from it and others not. I was moved to non request pricing without any consent.
Are you on the latest version of cursor, I’m not seeing the 2x requests with the latest version which I think is 1.4.5
Are you on the latest version of Cursor?
@Azuremis this may depend on the plan the user has.