My costs are exploding with cursor pro. I use mostly Claude Sonnet 4 (thinking) which seems to be the best model for most use cases.
I now compared to claude code pro.
Cursor Pro: 500 requests / month Claude (code) Pro: 45 every 5 hours => >200/day => +6000 / month
is my view right?
Difference with max Plan is even bigger.
Do I pay more than 10x for just accessing different models (most models make no sense)?
Edit: Cursor counts a normal sonnet request as 0.5 and with thinking 0.75. No information about max modes(?). So my calculation is not exakt but difference is nevertheless very big comparing claude pro and cursor pro.
I think also there should be a plan between cursor pro and cursor ultra as claude provides.
I think the latest update to the request systeme, where it became unlimited/rate limited, is to align with the Claude Code offer. You are no longer stuck at 500 request. But from my testing I seam to hit the rate limit quite fast.
@doczoidberg 45 with no tool calls, when tool calls happen, it may go to even 10 prompts every 5 hours, also most users would not use it for 24hrs so that calculation is misleading, pricing was good before and is even better now, as @TMRomain said, it was an update to get on par with other services pricing and to allow unlimited tool calls and background agents work in tangent
No regular claude-4-sonnet, I mean I can still do request on the model but it is super slow and hang on reading some short files. When I try same prompt with same files on VScode copilot it’s a lot faster. Now I have waited for like a few hours and came back to it and it seams fine.
Pretty much the feeling of everyone, what are the number of request before we get limited, after how much times does it reset. Is it counted per request to Cursor from the user or per cursor request to AI providers.
As Dan from Cursor already answered in other threads, I will rehash it here.
The actual rate limits should not be reached by most, unless you use a lot additional features like background agent etc, when eventually the ultra plan would become better match.
Dan also mentioned that there are sustained rate limit (longer period of time) and a burst rate limit (short period of intense activity). Basically burst would allow you to use more request in short time then continue with other steps (like reading docs or testing,..).
Technically its not really a difference how it is counted as it also includes tool calls and other new features.
If you are still able to send Claude 4 requests, but they are slow / hanging, this sounds like a bug and not a part of the new pricing!
We already have some fixes that should come in the next update that should help here, but if you’re seeing this a lot, feel free to post in Bug Reports
I understand that people would abuse the systeme if you guys gave us presise info’s on the rates limit, but indering/blinding us the regular customer for it seams unfair. Let’s say’s I code a lot a set day because of the sustained rate limit I might not be able to use the tools/model that suit’s me better for the rest of the hour/day, I wont be able to know when I can start to work with cursor again.
+I wont be switching to ultra because I have student discount, and even if I was paying jumping from 20$ to 200$ is a lot a intermediate plan would be great, or maybe the usage-base pricing is a good alternative
Dan mentioned in other posts that users can use Pro plan with usage based pricing. e.g. one user asked about 100 USD plan which basically would be from Dans answer a Pro plan + 80 USD usage based limit set.
I actually think this is a good idea in theory. I’m subbed to claude max and cursor’s former model didn’t seem appealing in comparison…now whether it works well in practice idnk
One thing I havnt seen much of yesterday is that summarising chat rubbish, that broke the flow and context every single time. Hope it doesn’t rear its head again, horrible feature.
