Questions about cursor's charging model and model calling strategy (gemini 2.5 pro, claude 3.7, and more)

Feature Requests
1

Dear Cursor Team,

First, thank you for creating such an excellent AI-powered coding tool. Many developers, including myself, rely on it to boost productivity. However, we are deeply confused and disappointed by recent developments.

Google has announced that Gemini 2.5 Pro Experimental (with 1,000,000 tokens) is now freely available in Google AI Studio. Yet, Cursor continues to restrict access to the full model—even for paying users—while pushing a paid model that offers an inferior experience compared to Google’s native platform.

Why Charge for What Google Offers for Free?

Google AI Studio provides free access to Gemini 2.5 Pro (including 1M tokens).
Cursor restricts free users and even paying users get a worse experience than using Google directly.
We understand charging for Cursor-exclusive features (e.g., codebase indexing ,tools,plan), but why monetize basic model access when the upstream provider (Google) makes it free?
Why No Freedom to Choose Models， The model called by the tool can be selected by the user?

Cursor forces its own model strategy, which feels unfair given Google’s free offering.
Why Do Paying Users Get a Worse Experience?

Directly using Google AI Studio or Gemini Advanced provides longer context, faster responses, and better performance.
If Cursor’s subscription is just “paywalling what should be free,” it harms loyal users.

Since Google offers it for free, let users call it directly (like other tools do).
:white_check_mark: Give Users Model Choice,The model called by the tool can be selected by the user

Allow switching between Cursor’s optimized models, Gemini 2.5 Pro, Claude, or local LLMs.
:white_check_mark: Charge for Real Value, Not Artificial Restrictions

Final Words
Cursor is a great product, and we’re happy to pay for real value. But if subscriptions just “unlock what’s already free,” it feels exploitative. We urge you to reconsider your model strategy and make Cursor more open and user-friendly.

— A Paying User Who Cares About Cursor’s Future

6 Likes
2

As a user I can see following:

  • Google offers it in its own tool for free to get people to use both the model and the AI Studio and eventually start becoming paying customers.
  • Cursor can not get it for free as it hasnt made the model itself and has to use the API where Google charges Cursor for usage.

This makes it logical that Cursor charges for it.
Asking a company to offer things for free where they already offer it for a reasonable price is a bit ridiculous.

Additionally you can use your own API key and get it that way cheaper or free but with other restrictions (tokens/minute not enough to even use 1M tokens as regular limit by google is 40 000 tokens/minute, so not even enough for a regular prompt with few files attached)

As for Cursor offering ‘worse’ experience. I wont go into judging if something is proper or not, many have here different opinions and experiences.

I will say that if you look at what features go into Cursor context its not just your prompt

  • .cursorrules file or User Rule
  • Users Project Rules that apply
  • Users Project rule descriptions that select rules
  • Users Chat thread (which is the main consumption of tokens across all your messages)
  • User configured MCPs with their tools listed
  • All tool usages incl. MCP that can retrieve a certain amount of info or need some info.
  • All attached or used files.
  • Also note that all input AND output tokens count together and not just your ‘files’
    etc…

Lots of parts come together and Cursor has to make a balance to allow multiple requests and responses in one thread.

Not sure why you misrepresent the situation. You can use Gemini 2.5 for free if you want, as described. Or use it as provided by Cursor at a price.

Also you are able to switch models, nobody prevents you from doing that. Or?

4 Likes
3

Unreasonable tool call charges

Each tool use (such as file retrieval, code analysis) is counted as an independent request, resulting in:
High hidden costs: users may not realize that these calls are continuously consuming quotas.
Experience fragmentation: tasks that could have been handled in one go with the native model (such as Gemini 1M token) are split into multiple billed requests.
Token limits are much lower than the native model

Even if Cursor uses Gemini 2.5 Pro, the actual available tokens may be much lower than 1M because:
The system automatically injects context (such as .cursorrules, project configuration) to occupy some tokens.
Tool calls (such as MCPs, file retrieval) may further split the context window.
Result: users cannot really take advantage of Gemini’s full long context capabilities, but instead pay more.

Is this what you think of as provoking a dispute? I am a paying user myself, and I am willing to pay, but you seem to be a member of the cursor team arguing for your own team? I didn’t start any dispute, but was actually discussing your model and how to improve the user experience. You made me feel like I was taking advantage of you by using the cursor, which is the most ridiculous thing.

5 Likes
4

If you want to seize market opportunities and respond quickly, just like other companies, and launch your product for free, then the cost of marketing should be borne by your own company, instead of arguing here about how much cost your server has borne or how much cost your model has borne. Is there anything wrong with you bearing the costs to seize the market in the early stages?

1 Like
5

Ok tool charges for MAX only are different topic unrelated to any model.
I agree that certain calls are less costly to Cursor than others. There are several threads on that where it is debated as you mentioned.

Yes its a good point that we currently do not have control as users of if we want certain parts to go into prompt or not as an AI decides that based on Cursors tested and prepared processes.

No i am not a team member and i said im not judging what is right or isnt. I posted my personal opinion how i see services. Also by reading your post i felt that you claim that cursor is charging for stuff that should be free like Gemini which was the main topic of your post and i disagreed with that only. Nothing personal against you :slight_smile: nor am I claiming you are provoking.

To perhaps mention, Im using 3.5 because i see from my usage that 3.7 just goes off the rails often (and that is not just in Cursor but also in other similar tools that use 3.7 and 3.5).

Sure market is important. But cursor has already paying users and needs to provide stability there from my point of view. So many people push new models like the holy grail and complain why it isnt in 5 minutes after Google announces stuff.

I tried 2.5 and it didnt do well what i need. But cursor apparently tested stuff and they saw a benefit to offer larger context.

Personally i do believe that Cursor is eating up part of the cost already as their customer growth pushes usage forward. But as I am not employee i have no real idea. Doubt however that its as its now already profitable as they have to continuously develop features AND some users clearly abuse the tool (saw some threads in the forum, wont go into details).

1 Like
6

Just to add one thing that came to my mind is. The limit that Cursor sometimes faces is also the AI providers request limit on API even for companies like Cursor who sure are on the highest tiers and maybe even above that on custom levels perhaps.

But it does happen that Cursor reports they are have rate limit issues in the UI as several users posted with screenshots.

That was seen more often when new models launch and everyone wants to use that model in what ever way they have access which eventually forces the AI provider to set stricter limits temporarily.

So even if customers pay for MAX but the AI provider cant handle the traffic yeah it fires back for the AI provider AND for Cursor.

Im not claiming that cursor is error free or perfect or etc. We all have likely good and bad experiences.

So i take back the word ‘ridiculous’ as i better understand what you mean :slight_smile: but didnt re-edit that post anymore.

1 Like
7

I also haven’t re-edited any of my posts, I don’t know why my first post shows as edited twice, but I have a backup of my original post locally. So I don’t care who edited my post

1 Like
8

The truth is only one if I pay the cursor because I’m paying and I don’t want to use Max and I want to use my own Google API for it even so it’s nerfed?
What is the rational cause?
I confess that I am not interested in explanations, but reasons for this to be done

Because the cursor doesn’t say the. Following:

Look at you using your API I charge you X and I won’t nerf the model I would pay but he doesn’t do that you know

Another point. Important you could say and but the tool apply is composer and agent would not be compatible

I’ll answer you again: I’M NOT INTERESTED IN THIS, I just want to use my Google API. On the cursor without her being nerfed. And because it doesn’t offer this, it’s a limited tool

Why limit the experience? Just to sell usage-based models? Ridiculous thing this

Unfortunately, not currently on the market. There is a more complete tool than the cursor I love it but if it had a similar tool with. The integrity as complete as his without nerfing models I would definitely migrate

And yet I say more I’m not interested in. Free unlimited requests

I just want not to be nerfed kkkk understand? Is that it

1 Like
9

You’re absolutely right. Cursor’s behavior is rapidly eroding its competitiveness in the market. While currently, compared to other AI-powered code editors, Cursor remains quite useful, I firmly believe that if they continue down this path, larger competitors will inevitably surpass them with superior alternatives.

As much as I’d love to see them improve, it seems they’re completely deaf to user feedback.

2 Likes
10

It’s fair but its 0.05$ price only depends on the need to not kill the 0.05$ price on the Claude 3.7 Max. Imo. Overall this does not seem sustainable and this price in itself is completely arbitrary and probably 100 times what the API costs Cursor. In a lot of ways it seems like a cash grab unfortunately. I hope this will change.

2 Likes
13

Lol someone censored my comment? Yikes

1 Like
14

Flagging My Post for No Reason , By saying I am going Off Topic
LOL Community

2 Likes
16

Exactly, and thanks @T1000 for quality posts, this one directly addresses a point I made to Cursor team around one month ago, leverage R1 to go away from Anthropic, and now Google, this should have been done before they released Cursor-like editors but we have time as Cursor is still the best, R2 is coming soon, I know the tooling capabilities are not on par with Claude/Gemini but here’s the great catch: R1/R2 can be finetuned by Cursor to use their tooling, distancing itself from competitors that can throttle requests or other harmful practices, and its just a start as it then can finetune models for example to directly address specific frameworks[…]

2 Likes
18

Editing my post and removing the solution post I chose means you don’t care about losing any paying users? And those loyal users? The person who edited the post, why don’t you just delete the post?

1 Like
19

While Gemini is free to individual users, Google is actually footing the cost for the API, and you would likely find it highly rate-limited in actual use. Also, requests will likely be used for training by Google for future models.

With Cursor, at the scale we operate at, we still have to pay Google for using the Gemini models. We do ensure that none of your data is ever stored or used to train Google models. However, I understand this seems counterintuitive.

As a solution, you can always provide an API key inside Cursor to use your own key and not be charged within the limits of your usage. With a Pro plan, all Cursor features can be used with an API key at no additional cost!

If Gemini 2.5 were free for us, we’d add it as a non-premium model and let everyone have unlimited usage, but unfortunately, this is not possible.

@oooooono Regarding your edit, I can’t see what you are referring to in the edit history of the message, but I apologise if your selected solution was removed for whatever reason!

4 Likes
20

100% agree this is what they need to be focusing on. If Cursor wants to keep the advantage in the AI IDE space, they need to be leaning into models like R2 and the upcoming open-weight model from OpenAI and fine-tuning those models to make the “cursor models” work impeccibly with the Cursor workflow.

Because it’s only a matter of time before Google launches their on VSCode fork with native Gemini support, and since Google is already ubiquitous, they’ll basically sweep up the entire enterprise market.

1 Like
21

Does this include 2.5-pro MAX is I’m using my API key?

Right now the UI still says $0.05 per-request, while my usage does list the calls as “User API” the disconnect makes it difficult to know whether or not I’ll be charged.

1 Like
22

Hey, if you enable a Google API key, Max usage would be charged only to the key, and would not incure any costs within Cursor.

You do however still need a Pro subscription, as we use some smaller models behind the scenes that are included with a Pro subscription.

1 Like
23

I’m a pro user and I have gemini-2.5-pro-max[key] enabled, why do I get the message

Request failed with status code 404: [{
  "error": {
    "code": 404,
    "message": "models/gemini-2.5-pro-max is not found for API version v1main, or is not supported for generateContent. Call ListModels to see the list of available models and their supported methods.",
    "status": "NOT_FOUND"
  }
}
]

image
image708×104 2.09 KB

requestid：552cb6ef-b6ec-4c70-a1cd-f57241832449

image
image814×142 4.41 KB

24

pro max was working just fine on my API key for the last 2 weeks, but as of last night is failing

all other gemini-2.5 models send successful API requests

Request failed with status code 404: [{
  "error": {
    "code": 404,
    "message": "models/gemini-2.5-pro-max is not found for API version v1main, or is not supported for generateContent. Call ListModels to see the list of available models and their supported methods.",
    "status": "NOT_FOUND"
  }
}
]

Version: 0.48.7 (user setup)
VSCode Version: 1.96.2
Commit: 66290080aae40d23364ba2371832bda0933a3640
Date: 2025-04-01T23:50:21.645Z
Electron: 34.3.4
Chromium: 132.0.6834.210
Node.js: 20.18.3
V8: 13.2.152.41-electron.0
OS: Windows_NT x64 10.0.22631