Cursor team just wants to keep things secretly priced, I guess…
Also @danperks, if we revert to the old plan, do we get unlimited slow requests + 500 fast or just 500 fast?
@danperks firstly thanks for your reply.
- On the last update (I’m assuming this is what caused it) my model got changed to thinking and non thinking was removed from my shortlist.
- I didn’t realise this until I started getting rate limited and noticed.
- Now my burst limit is gone and it seems to build so slow it may aswell not be there
The issues I have with the new system vs the old x requests per prompt is that now some of my prompts are:
- No transparency on where we are at with the rate limits.
- Some prompts I use are now 20x the cost as I like to set things up and run a large prompt. I understand that it’s fair and in line with Claude’s cost, however this wasn’t clearly advised before it changed and I was blindsided by the change.
- Auto is just bad so it’s not really a viable option. I’ve only ever had sub-par results with this.
- There’s no $40 plan for 2x which may honestly be enough for me, but it jumps up to $60.
- I don’t use my limits every month, some months I use Cursor much more. This means now if I upgrade my sub I’m paying extra for something I don’t use.
- The period where I couldn’t see my cost tricked me into thinking I hadn’t started using it yet, which meant I could work on smaller, less productive projects which I would not have otherwise.
I could probably work with these issues and optimise my prompts, etc. But even this is made hard because I can’t see the cost of a prompt straight away.
I need to use a prompt, wait minutes for the dashboard to update, check the cost, use another, see the difference etc.
Because there is no real clarity around any of the issues, it’s now difficult to know what we are using, how much money we are spending and why we are being limited.
I have a Pro Plan and using Claude 4 I hit the rate limit somewhat quickly. I was a bit surprised, but then I switched to Claude 3.5 and since then never hit the limit again, even though I’m using Cursor pretty much non stop 10 hours a day. I now clearly got much, much more out of it than if I was using the old pricing plan. Not even comparable.
Now, I know that my use case doesn’t fit everyone. For my work, surprisingly, Claude 3.5 does a better job than Claude 4 that in my opinion tries to do “too much” too often. I find 3.5 to be more “straight to the point” and efficient.
Anyway, all this to say that I feel that the new pricing can give you much more value than the old model, however, @teknoman nailed it in my opinion. Cursor should add two usage bars to display the limits behavior so users can anticipate and adapt. It’s more an issue with the UX than anything else. The old system, even if it could be unfair, was at least very predictable. And predictability is of the utmost importance when you’re trying to run a business.
Please consider adding a bit of feedback for the user with regards to limits. You mention two types of users, and the second (those that are familiar with usage-based pricing) probably has a good grasp on what affects cost, but your average user (the former group) has absolutely no idea what is more expensive.
The current implementation provides zero feedback to the user. Imagine going grocery shopping, but you have no idea how much any of the items cost, or how much money is in your bank account. You know you’re getting paid, and you have some money in there, but the only feedback you get is when you get to the counter whether there is enough money in your bank account. That’s it. It’s very hard to be a smart consumer when you don’t know the amount you have and you don’t understand how much something costs.
I’m NOT asking for highly technical information. Perhaps just a colored progress bar. Green, you’re in within your local limit, yellow you’re in your burst and you can see it actively dropping, red you’re on auto-only mode. Then I’d know, oh, I only get roughly X % back a day, and I know I have a big day coming up, I’d better back off. In short, it allows me to be a better consumer. I can budget, I can make educated decisions about when I need a more powerful model and how much that’s going to “cost” my available budget. It even shows me when I’ve recovered my local limit. You could add tooltips to this explaining it, and explaining how to reduce and save budget.
It’s just an idea. I know you have costs, and I respect the incredible value you bring. But I can’t responsibly use it without more information.
On a separate note, it would be useful to know what model was chosen when in auto mode. That way if it’s doing a good job, I can begin to make rational decisions when I do force a model rather than just blindly going, “well, this is a tricky problem… and X model is the best right now, I better use the most expensive”.
Just things that I think would qualm a lot of the noise. Right now we just don’t know, until we get up to the counter and find out NOPE, no money in the account.
You’re absolutely right, they’re quietly rate-limiting paying users without disclosing the exact thresholds. That’s not just shady, it’s anti-consumer.
Here’s a screenshot from today (July 2nd) using Claude 4 Sonnet MAX mode. Look at the frequency and costs per prompt: some interactions are $0.06, $0.13, $0.29, even $0.71, for a single request. I hit ~$4 in under 30 minutes with no clear warning or “limit” alert.
Where’s the transparency?
The current system seems to use:
$20/month cap (okay, fair)
Local limits (4h) — No one knows how many requests you’re allowed.- Burst limits (24h) — Again, not defined.
No dashboard to track usage per model or reset times.
And here’s the info we’re still missing:
| Model | Normal Mode | MAX Mode |
|---|---|---|
| Claude 4 Sonnet | ??? req/day | ??? req/day |
| Claude 4 Opus | ??? | ??? |
| Gemini 2.5 Pro | ??? | ??? |
| R1 0528 | ??? | — |
| GPT-4.1 (OpenAI) | ??? | ??? |
| o3, o4-mini (OpenAI) | ??? | ??? |
| Auto (blended model) | ??? | ??? |
All we’re asking for is a simple table like above—fill it in, be honest, and stop gaslighting power users. Cursor’s strength is in dev trust—this hidden throttling is a direct breach of that.
If the limits are strict due to cost? Fine. But say it clearly. Don’t hide behind vague “usage-based” billing while draining people $0.70 a pop on MAX mode prompts.
Right now, it’s way too f* expensive to use the application.
Yes, totally agree. Cursor pro may not be bad, but the Cursor team is giving us every right to assume (and therefore, to say) that Cursor pro is bad, because we don’t know how good it really is. I’m probably sticking to the free plan. Definitely anti-consumer.
You “really like” a pricing model that hides rate limits, burns through your $20 in 30 minutes, and gives you no visibility into usage?
Either you’re not using the same Cursor we are… or you’re a dev team intern moonlighting as a hype man.
Let’s be honest, no one’s mad about paying. We’re mad about paying blindly.
@danperks Hello my friend, some people have shared quite constructive criticism on this topic. So why does it say “This topic was removed from the list” next to the thread title? Instead of considering the constructive feedback, why do you choose to hide the discussions?
Just like the other said, just 4 prompts I hit the rate limit. Never happened before but after I posted message here, I got rate limit. Someone must be watching lol…Claude Code calling…