Honestly you guys are doing the worst PR job I’ve ever seen.
I feel like canceling my subscription just to avoid giving you any more money.
11 Likes
well it was fun while it lasted, rip Cursor, raised hundreds of millions then cut usage allowances in half
10 Likes
Hey,
After reading your “sorry-not-sorry” announcement, I feel like I’ve been left with a really bad deal as someone who paid for an annual Cursor subscription.
You’ve formally responded to criticism about transparency, but you haven’t actually addressed the core complaints about rate limits.
In fact, it feels like you’ve doubled down: the previous “short burst”, “local burst” rate-limit narrative is gone, and now all Cursor offers is $20 of API usage per month?
You’ve added a dashboard to track API spend in USD, but there’s still no way to see (either in Cursor or the dashboard) if we’ll ever get more in the “local rate limit pool.”
The old system offered 500 fast requests (plus unlimited slow requests); now, it’s just $20 of API usage, which feels like a huge downgrade, especially since the non-MAX versions have a much smaller context window (120k vs 200k).
At this point, why not just use Anthropic’s API directly and pair it with RooCode or Cline?
Other options are simply clearer and more generous. Anthropic’s own Claude Code gives you “unlimited” (within rate limits) requests with the full Sonnet context window for $20/month. You get about 30–40 requests every 5 hours, and the rate limit is clearly communicated with a warning message.
GitHub Copilot Pro offers 300 requests for $10, with a 128k context window (and 200k support is coming soon) for Sonnet.
AugmentCode gives you 600 requests for $50, with full context enabled. It uses Sonnet 4 as the base model, and you can’t choose any other.
Given all this, I honestly don’t see any reason to keep my annual subscription.
Is it too late to request a refund for the remaining part of my annual subscription?
This whole experience has been a huge disappointment.
39 Likes
when using auto mode why it cost?
About tokens limit and context windows.
I’m confused with why I limited with 120k but real amount is > 2kk ))))
1 Like
@Mental it says included. its not charged extra.
@danleech 120k is input, ~2m+ may be cached from whole session. See “Token” view on that page top right to see what the input vs cache read was
I paid for annual Cursor subscription too, could you please share how to cancel the plan?
2 Likes
Ever since this change, the max toggle in models docs page no longer works. Also I find it bit difficult to understand what models exactly “Auto” routes to. It says premium models.. which means? Does this include sonnet 3.7 for example? Or o3-mini? It would be nice to list the models, perhaps add another icon to the models list in docs to see which are available in auto and which are not. Also the pricing toggle is gone from models docs page, now I dont know what the api costs are and have to lookup each model on my own… On a positive note, I find this page explaining rate limits quite well, honestly better than any community or blog post. Also thank you for providing subscription usage in dashboard, it is great.
1 Like
Totally agree—Cursor’s “explanation” post skirts the real issue: opaque rate limits and shrinking context windows. For $20/month you still can’t see how many fast requests you get. Services like Claude Code ($20 for true unlimited within limits), GitHub Copilot Pro (300 calls for $10, soon 200K context), RooCode/Cline with BYOK, or AugmentCode (600 calls on Sonnet 4 for $50) are far more transparent. We should request a refund and moving to a tool that actually shows what you pay for.
2 Likes
Going back on unlimited and then changing the pricing pages to cover it up for Ultra is unacceptable. Its clear they just rolled out the new plans without thinking or testing adoption on segments of users to validate actual consumption to see if it was viable first.
I have been on a paid cursor subscription since October 2024. But now I have to say goodbye
12 Likes
@condor Could you share the screenshot where I could find it?
I tested with default model and it seems it’s ok.
It might depends on project because on a one with PHP I get >1m from the new chat.
Auto does use claude when it is neccessary. The issue is people were wanting to use only sonnet thinking max and nothing else. They weren’t on auto. The pricing has always been ckear the issue is people keep trying to abuse the system. Your project does not need claude models for every request. No project needs a certaian model.
1 Like
Is this a clarification or a price change again? Is the headline fake?
My email to cursor:
Deeply confusing posts on the forum have led me to believe this organization is changing the plan I signed up for with hidden messages and causing A LOT of confusion. I’m unsure what is even unlimited at this point… I would like some refund for my usage during this “transition” to whatever is being mentioned.
Not happy with the company’s PR right now. Please consider your reputation as a brand before making widespread changes like this.
Thanks
They are never willing to let us know which model exactly is used for auto
What would happen with current annual pro subscriptions? Is there any kind of distinction in the way requests are handled?
In my case, I paid for an annual subscription months ago, which is still valid, but I can’t clearly understand the number of changes in pricing and business models. It’s actually as if I had paid for a year for a certain type of service that has completely changed in the middle of the process, several times in fact. Last month I paid about $150 in tokens in addition to the pro subscription, and I didn’t use it very much. I mean, I stopped using it because, looking at the dashboard, some requests seemed really high: $1, $2, etc.
This led me to go back to using alternative products, which actually work well and at least have a clearer or less “changing” or more predictable policy.
Honestly, I’m a little afraid to use Cursor because I don’t know how much the request will cost, or if it’s included in the Pro plan, or if there will be a charge, etc. And it’s a shame because I think the new versions incorporate some very good things and the product has progressed a lot.
I think there are some very basic things, beyond the dashboard, such as being able to see the cost, expense, progress, and so on, incorporated into the IDE itself, which would be essential. Previously, Cursor Stats, an independent extension, worked, but it doesn’t make much sense to have to use something “non official” to know how much we’re spending. This is something that Copilot, Cline, Roo, etc. already implement. It seems pretty basic and logical to me.
1 Like