I’m trying to get a clear understanding of Cursor AI’s pricing plans, especially regarding Premium and Agent models. I have a few questions about how pricing works beyond the free limits, and I’d appreciate any insights.
Premium Models
From what I understand, Premium models (e.g., Claude-3.7-sonnet, GPT-4o) come with 500 free requests per month under a Pro or Business plan. If I want to continue using them without limits after reaching the 500-request cap, I can enable Usage-based Pricing in Settings.
My question:
If I enable Usage-based Pricing, will I be charged per request once I exceed 500 requests? (For example, Claude-3.7-sonnet is listed at $0.04 per request—does that mean I will be charged this amount for every additional request?)
Agent Models and Tool Calls
If a request includes more than 25 tool calls, Cursor will ask if I want to continue. What happens if I continue?
If I use MAX mode (Claude-3.7-sonnet MAX), each tool call is charged separately—is there a fixed price per tool call?
If I do not use MAX mode (e.g., just Claude-3.7-sonnet in regular Agent Mode), do tool calls incur an additional cost?
If tool calls are charged outside of MAX mode, does the pricing follow the same Usage-based Pricing logic as Premium models? (For example, if I’ve exhausted my 500 monthly requests and have Usage-based Pricing enabled, would I start paying $0.04 per request for GPT-4o?)
I’d appreciate any clarification or official guidance on how Cursor calculates these charges. Thanks in advance for your help!
as i have some experience using Cursor I will try to answer your question but if my answers are not matching your requirements Cursor team may need to have a look. The info is per todays docu on Cursor website, note that I do not guarantee the info provided.
Usage-based Pricing
Yes after you use 500 fast requests (premium) you would be charged the USD price in the model table for usage based pricing (see docu).
However some models are only available on usage pricing (MAX,d o1 ,.. see docu)
Some models are cheaper than the regular $0.04 price, see docu.
If a request includes more than 25 tool calls , Cursor will ask if I want to continue
Each continue is charged x requests as per docu depending on model. (some 1, some 2 and some 1/3 request or the USD value if on usage based pricing after using up fast-request in package or for models that only exist in usage based pricing)
If I use MAX mode (Claude-3.7-sonnet MAX), each tool call is charged separately—is there a fixed price per tool call?
yes $0.05 per tool call as in docu.
If I do not use MAX mode (e.g., just Claude-3.7-sonnet in regular Agent Mode), do tool calls incur an additional cost?
No, tool calls for regular (non-MAX) models cost no extra charge with limit of 25 tool calls per prompt included.
If tool calls are charged outside of MAX mode, does the pricing follow the same Usage-based Pricing logic as Premium models? (For example, if I’ve exhausted my 500 monthly requests and have Usage-based Pricing enabled, would I start paying $0.04 per request for GPT-4o?)
Not charged per tool call and includes 25 tool calls. Continue charges same as a new request as it starts process new.
Each non-MAX request is charged on Usage-based pricing as per price table in the model list, see below.
To clarify: most models are 1 request or $0.04, some are 2 requests or $0.08, some are 1/3 request or $0.03 and some are purely usage based price as listed in the docu.
When a user enters a prompt and presses Enter , does Cursor immediately count it as one request , or does the request only count if the user accepts or applies the generated code?
If a user rejects the generated code and asks for a revision, does that count as a new request?
In Agent Mode, when executing commands in the terminal, does each command execution count as a request, or is it included in the initial request that triggered the action?
If a request involves multiple tool calls(e.g., generating, editing, and running code), is it still counted as one request or 1 tool call, or does Cursor charge separately for each stage of the process?
When a user enters a prompt and presses Enter , does Cursor immediately count it as one request , or does the request only count if the user accepts or applies the generated code?
Requests count (prompts you write into the field and send to Cursor servers), not accept/apply/reject.
As mentioned the tool counts (apply change, edit, MCP,…) count on MAX model.
Basically when you submit a request it is processed by Cursor and forwarded to AI provider (Anthropic, Google, OpenAi,…) which has a cost for Cursor, therefore even if you dont use the response it counts.
This also applies to all modes where you use premium or non-free AI models.
Free AI models included in subscription don’t cost but may be less powerful, see list.
If a user rejects the generated code and asks for a revision, does that count as a new request?
Yes, every prompt you write and submit or click on continue after 25 tool calls on non-MAX models counts (per model prices as request or USD).
As mentioned 3rd party AI providers charge usage to Cursor which Cursor charges to your account.
In Agent Mode, when executing commands in the terminal, does each command execution count as a request, or is it included in the initial request that triggered the action?
Terminal is a tool call and charged as such (see previous response above, cost depends on non-MAX or MAX model)
25 tool calls included in non-MAX models.
Not all models support Agent functionality, list shows Agent as checkbox when model supports it.
If a request involves multiple tool calls(e.g., generating, editing, and running code), is it still counted as one request or 1 tool call, or does Cursor charge separately for each stage of the process?
This remains the same as before, on non-MAX models each request (prompt you write) costs requests per model list.
The tool calls (search code, edit, re-apply, MCP, terminal etc…) on non-MAX models that have Agent function are included up to 25 tools per request.
Each new request charges the request and includes 25 tool calls maximum.
MAX model costs per request AND per tool call.
Models that have no Agent checkmark in list can not use tool call.
Perhaps to add:
Non premium models are cheaper or free, per list but also less powerful.
Each subscription includes 500 fast premium requests, after you use those fast premium requests the requests go into slow queue. The more you push into slow queue the slower it gets (fairness to other users).
You can enable Usage based pricing for unlimited amount of metered fast request pricing and that is accumulating after using up fast requests OR when you use models that are only usage based charged (o1, MAX,…)
I suggest starting with the free trial and getting used to writing prompts, learning how to write rules, using Cursor features etc.
Docu has quite good info about features.
Start simple and expand the more you know about Cursor and the AI models used.