even the forum is club fan, the real ppl that are getting the money of the vibe coders really dont care about this big fan club, the only person is trying to hide what is happening is the mods and see “@0xHACKS Sorry about that - I have relisted your thread!” so really i know in reality dont matter if we do 1 emotional post or mass post but it will start smth ik it will start smth the things will not stay the same.
You don’t have to wait until next year. You can apply for a refund and then switch to other AI tools
I got a usage warning earlier this week estimating I’d hit limits by July 21 — but after barely using Cursor for a few days (I was on vacation), I came back today (July 17) to find myself fully limited out. What’s frustrating is that this month alone, I’ve upgraded from Pro → Pro Plus → Ultra just to keep working, and now even Ultra is locked.
The system’s usage projection and actual lockout don’t seem to match, which feels super opaque. I also reached out to support, but seeing this thread, I’m realizing it’s not just me.
If anyone has tips on how to push this with support or if anyone from Cursor is watching here, it’d be great to hear how you’re planning to address this pattern — especially when people are actively upgrading but still hitting surprise blocks.
Exactly. The real devs making money off Cursor aren’t sitting in this fan club thread, they’re cashing checks. Mods trying to bury threads won’t stop people from noticing what’s happening. One post won’t change it, but a wave will. This won’t stay quiet forever.
I took my current billing cycle’s usage data directly to Claude.ai to ask the Sonnet 4.0 model about it. There’s got to be a bug with how it’s reading from the cache files because the cache calls are so significant higher than the input tokens.
The drain is most certainly related to how Cursor interacts with Claude 4, and it’s inefficiencies. I took my usage information for this billing period directly to Claude.ai and asked it what’s going on, here’s what it says:
You’re absolutely correct! The cache usage pattern for Claude-4-sonnet-thinking is completely abnormal compared to the other models. Let me analyze what you’re showing:
Claude-4-sonnet-thinking (problematic):
- Cache Read: 15,982,739 tokens (90% of total usage)
- Actual Input: 48,104 tokens
- Cache Read to Input ratio: 332:1
Other models (normal):
- Gemini: 21,481,939 cache reads vs 2,201,031 input (10:1 ratio)
- Auto: 4,674,816 cache reads vs 3,232,908 input (1.4:1 ratio)
- Grok: 206,144 cache reads vs 632,679 input (0.3:1 ratio)
The Claude model is reading from cache at a rate that’s 20-100x higher than the other models for the same types of requests. This suggests:
- Broken cache invalidation - The Claude model may not be properly recognizing when cached content is still valid vs when it needs updating
- Over-aggressive context inclusion - It might be pulling in your entire codebase context even for simple queries that don’t require it
- Cache thrashing - Repeatedly reading the same cached content multiple times per request
- Inefficient prompt engineering - The model might be structured to always request maximum context rather than being selective
This isn’t about your
.cursorignoresettings - if it were, all models would show similar patterns. This looks like a specific implementation issue with how Cursor handles caching for the Claude-4-sonnet-thinking model.You should definitely report this to Cursor support as a potential bug. The other models are demonstrating much more reasonable cache usage patterns, so this seems to be a Claude-specific optimization problem in their integration.
1 Like
Nobody buys annual subscription. Ai is reshaping the tech industry every week. What you are using right now might be obsolete within weeks.. Go with monthly..
5 Likes
I was on the $60 plan, and my “unlimited” quota was gone in literally 10 days. What kind of unlimited plan runs out in 10 days??? With the old $20 plan, I used it for almost a month and had more than enough. That plan only ran out four days before the end of the month. But now, with the $60 “premium” plan, I barely got 10 days, and now I have to look at the rate-limit message for the remaining 20 days. What’s the point of paying $60 for such a short period?
If you make a new chat tokens will be reduced.
nobody as in “you”
because I’m in my second year of annual billing
That’s why I went with Cursor as it’s an abstraction over the underlying models with the promise of “better models, same billing”
The guy expresses his frustration, and the @danperks just responds by saying he used $150 on a plan that only covers $20 — basically throwing it in his face.
Worse than that, they exposed a user’s private usage data here in the community — not just the original poster, but also others who commented.
You can’t get upset at users for being vocal — the mistake was yours, Cursor.
You messed up with misleading marketing and basically sabotaged everyone.
Now, it’s your turn to listen silently to the wave of dissatisfaction you created.
8 Likes
Argh - Many months of Cursor usage and now with lower usage than normal this month im over my plan limit 10 days into the month.
This experience is forcing a look to other options in market.
1 Like
@danperks Just checking if we can get a refund or reset of our usage when this bug is fixed? Cache usage if off the charts for me and just seens wrong.
Forced into auto mode and trying to do similar work (painful), completely different cache usage (and token usage) for same work, same codebase, same chat thread etc
1 Like
@danperks
Hello there 
My final offer is simple: raise Ultra to $250–$300 per month and remove all usage limits entirely. If you want committed power users, price clarity and predictability matter more than theoretical fairness. Ultra at $200 with constraints isn’t working for many of us, it’s equivalent to being on Pro+ with token surprises. At $250–$300, with no limits, you’d actually deliver on the ‘power user’ promise. No more bait‑and‑switch; just premium value. That’s it, that’s my final offer.
The AI landscape is moving fast, and unless Cursor starts showing reverse loyalty to its power users, many of us will move on. Better options are around the corner. It’s only a matter of time.
3 Likes
The cursor is dying, and your users will soon be snatched away by other competitors.
1 Like
Cursor has become extremely expensive to use. You have no way of knowing when your “extended” agent access runs out. Massive liability - especially when models can go “turbo” sometimes making a whirlwind of changes. Looks like I will be cancelling my subscription.
Sorry Dan, but a lot of users feel like the rug has been pulled from under them. One minute users had unlimited agent usage, then, without warning, this changed to an ambiguous number that is not shown on your dashboard. If I hadn’t been checking daily, I could have run up a HUGE bill.
“By introducing this new pricing structure, we hope to be able to continue offering great usage for users on every plan”
And your way of doing that is making sure loyal customers end up paying considerably more than ever before. Even with the old plan, a day working with Cursor would work out around ~$8 - Now it can rack up $20 per day with LESS use!
Sadly, I’m already looking for an alternative - all my trust in Cursor has gone.
3 Likes
I’m using Ultra, and although there’s no request limit, my Sonnet 4.0 usage and others were recently suspended. I’m paying around $400-$500 separately, and there are still 8 days left before my subscription expires. BUT THERE ARE LIMITS COMING. SUCH NONSENSE IS UNACCEPTED. WE ARE BEING FORCED TO USE DISGRACEFUL GPT VERSIONS.
1 Like
You should also check your payment and card details. Extra charges are being taken. AI is responding to support emails. While I was in love with Cursor, it’s turning into hate.
1 Like
Why is there a mid-month charge on a monthly plan? Why would I set a monthly “LIMIT” if it jist charges whenever?
Cursor is dead. Move on to Kiro. This is a ■■■■ show gone over board. GG!
1 Like