For those who can only spend $20 per month on Cursor. Refer to the article!
Im assuming this only applies to the old model of 500 requests per month?
2 Likes
Now Cursor no longer limits toolcalls, so even if you use 500 quick requests per month. You still save a lot of requests. Why? When you use 1 request, Agent will call ai_interaction_tool, from now on, you only need to use ai_interaction to chat with Agent, and no matter how many times you chat with ai_interaction, you still only use 1 quick request → There are still 499 quick requests waiting for you to use (But this does not apply to MAX MODE because MAX MODE will consume quick requests for toolcall)
But we are also charged by tokens now vs being charged by ‘request’.
If we opt in to the old request based charge, its still limits to 25 toolchain calls.
Atleast this is my understanding from the very dubious and vague changes to pricing that have happened.
2 Likes
I don’t think if you go back to 500 quick requests, you will be limited to 25 toolcalls. About the current token charge. I said it in the article. Using toolcall to chat will save your rate limit to the maximum, if you don’t use MAX MODE, I SURE you will never hit the rate limit if you use this tool.
I have been using ai_interaction since I created it. And now, I use Claude 4 sonnet MAX every day without ever hitting the rate limit. The first time I hit the rate limit was because I used Claude 4 Opus for 2 consecutive days and chatted a lot, simply because Opus is very expensive. Claude 4 sonnet MAX is cheaper so I have never hit the rate limit.
I believe it’s clear that, now for the old plan, each request is limited to 25 tool calls within the 500-request allowance. Have you tested this?
If you exceed 500 requests, even with the old plan, charges are based on compute usage, and there might not be a limit on tool calls.
For the new plan, everything is based on compute (API cost), so I assume there isn’t much difference in token usage between a tool call and a regular request.
1 Like
Can someone verify this or explain how this works?
You don’t know because you don’t use it. Definitely more economical than real request.
If I use real request like you. I hit rate limit many times.
2 Likes
Yes, it is better but it still isn’t unlimited, it’s going from about 5000 induvidual requests ~(500*10 are used per chat) to 12500 (500*25)
1 Like
I can confirm I am back to 25 tool calls
3 Likes
I confirm it too, opting out and using old plan = 25 tool calls
2 Likes
Please guide me how can i opt out and use the old plan, this new pricing plan is so frustrating, Its only been 10 days since i bought subscription and its already saying you hit the rate limit and switch to auto.
Dashboard on Cursor’s website → settings → advanced settings. Just make sure not to click on delete account button.
Also, opting out is not available if you have student free account.
And they might count those requests you’ve already made against your limit. I made double sure NOT TO USE any requests on new plan before I opted out.
No changes are needed. Someone has calculated that under the new plan, 20 dollars can request the following number of times (non-MAX mode): Claude4 (Sonnet) is 556 requests, Gemini 2.5 is 889 requests, and GPT4.1 can even use 1000 requests.
I tested how much it costs to use a cursor to send claude-4-sonnet-thinking to solve a simple problem. If it is $20 a month, I can only use claude-4-sonnet-thinking about 50 times, which is very different from the previous billing method (225 times).
3 Likes
Is every tool call counted as one request when opting out?
1 Like
In the case that I used cursor and didn’t enable max mode, it only took 10 days to say: cursor has saved you 120dollars in API fees. Now I can only use the sonnet4 two or three times every few hours, and I can only use gemini2.5pro and occasionally use o3.
1 Like
is that for the rest of your billing month? this new plan has really put a chill in how I use it because I’m afraid and have no meter to tell me how close I am.
so you can still use the premium models, but at a very throttled rate?