How to disable Cache Write and Cache Read?

Sorry but I do not have any information of that kind and this thread is not about that topic at all. I will add that the billing was announced as change from requests to token based billing.

Feel free to contact [email protected] for billing related issues.

You are aware that hi@cursor is a bot named Sam that responds and is not a human and struggles to provide real answers? Not a great answer / resolution to this.

I’ve seen hundreds of other cursor users (who have posted so far, probably tens of thousands are experiencing this). All of them who with pro or pro+ were able to work all month onthe project they have been working on without hitting their limits. Pro accounts are running out of credits in 3 days and pro+ accounts are running out of credits in 7 days.

It really seems like Cursor as a billion dollar company doesn’t care about its client. People aren’t going to pay 70x more than they did before for software, I see a blood bath of people leaving coming, and I’ll help lead the charge. I used Cursor every day, but now I won’t be.

No its not just Sam answering on email. Just today I saw staff responses on the email. But if you have an issue that was not sufficiently answered by email please create a separate bug report and mention that you already contacted the email address so it can be escalated.

I would pose this thread is related to it. Because people are being charged 70x what they were before, they want to disable the cache read/write function to avoid being over charged. What I’m discussing is the underlying cause of why people are trying to disable cache write/read.

Again Windsurf, Claude Code and others are not processing these unreasonably high cache write/read tokens, it’s just Cursor. Additionally, Cursor in Ask mode works perfectly without the cache read/write errors, it’s just agent mode where this happens.

There were changes and Cursor Team has acknowledged issues with that.

What I explained above shows how the cost would have been 4x more with cache read disabled. I don’t have other info, and the other threads have been addressed by Cursor Team.

Can you link me to where the Cursor team has acknowledged this, it would be great to see and I can’t find it. thanks

The latest is likely this one.

Thank you for providing the link. In the article it says “Based on median token usage, the Pro plan currently covers about 225 Sonnet 4 requests”. I got less than 60 uses before my plan usage was used up.

On the pricing still up on Cursors website
they clearly state:

This is why people feel mislead directly. Their newest post says only 225 sonnet 4 requests, when their live pricing says 675 sonnet 4 requests! I’m only get 60 sonnet 4 requests on Pro+ before using up all my credits, that is only 1/10th of the 675 sonnet 4 requests the pricing says I paid for. What’s more interesting is the cache read amount you broke down and explained was exactly 10x rereads of the cache write.

Sorry it won’t let me edit. Correction I meant to post 675 Sonnet 4 requests with pro+, not 225 sonnet 4 requests with Pro. My apologies, I meant the 675 sonnet 4 requests where I only get 60 (1/10th of the amount listed).

It says median user usage, as in average. However as Cursor does not count requests, this is an approximation since the actual number of requests depends on resources used (model + context & how its used).

Please use my link to see how you can optimize the usage and how your usage is affected by context attached.

There are many users who do not exceed Pro plan.

Some users do due to following

  • Intensive usage of AI (using heavier models, thinking models which consume more tokens, using AI much more than average user, … )
    and/or
  • Higher context (more tokens)

This doesn’t explain why the pro+ plan, with the same use on the same exact project/project size used to last all month, and now its running out in 7 days for me and everyone else. If that was the way it operated, why before did a pro+ last the entire month with same usage and same project? Again I only got 60 requests vs the 600 requests I was getting before this update, on the same project.

Its apples to apples comparison of usage between the same project, before and after. Hundreds of people are also complaining about this. Clearly there is an issue, where the amounts of request we used to get on the same projects are 1/10th what they used to be.

I’m using cursor just like I’ve always used it where its always lasted all month. Switching my request style, to meet a system that is using 10x more usage per request is just putting a bandaid on an open wound. How about Cursor fixing the open wound issue of 10x bleed rather than telling people to find creative ways to use 10x less usage than they were before. I used to get the correct amount of 675 requests per month, with the same project, project size and Claude 4 Sonnet model.

Do you have such a specific case? I would like the Cursor Team to look further into it.

I saw one screenshot you posted in another thread but not sure if its yours.

Specifically about requests from last 4 days to investigate. If you have a Request ID with such high token usage change it would help. (perhaps from chat history)

Thank you. That would be amazing to have Cursor team look into this. I’ve seen many people reporting that their plan that used to cover all month usage on the same project is running out in 3-4 days for pro and 7-9 days for pro+ users.

Cursor admitted today that they are baiting and switching annual users who paid for a amount of requests per month model, as I showed you the pricing site still shows a you get x requests per plan model. However cursor said it no longer a per request model it is a straight API fee model. Which is the bait and switch I was talking about earlier, customers who signed up annually based on processing by requests are being switched out of their contracts. This is exactly what I described, Cursor no longer holding to their advertised request usage promises.

I’m filing for a complete refund for all my cursor bills from my bank directly as a result of the bait and switch and publicly notifying the community about this.

Many people are complaining about exactly this:






















1 Like

@Illuminationx please no screenshots, nobody can message those users back or find those locations in forum.

I would like to forward specific Request IDs, or even log usages of last 4 days that would show such unusual counts. Would need to be able to contact the user.

Happy to jump on your requests if you have unusual high counts of last few days or if anyone has a specific post showing a high Cache Read count.

This is really about specific cases that can be checked about counts to verify them and not as much about total change from old plan to new plan as thats a separate thing to review.

For any refunds you have to contact [email protected] as those are not handled in the forum.

100’s of people are reporting the same issue as me, can we acknowledge that? I was looking for you to comment on dan perks saying that Cursor is no longer measuring usage by number of requests to the AI but by direct LLM API costs and switching the x amount of requests model that annual subscribers signed up for. It is illegal to switch the contract of usage on someone after they have paid for that contract. Hundreds of people are all complaining about getting a fraction of the usage they got before the update, likely tens of thousands of people are dealing with it as well.

Direct Link of Dan Perks saying this:

Can you acknowledge the reason people are getting changed 10x more, is because cursor is only providing 10% of the AI Model requests that everyone was getting a few weeks ago?

Again Cursor’s current pricing still shows the X amount of requests per pricing tier model, while they are charging by LLM API usage instead, which is why people are running out of credits in 3 days on Pro and 7 days on Pro+. This is a joke. The inability for Cursor to directly acknowledge the issue makes it seem like they tried to hide this new charging rate system, within the new pricing changes as if they are just the new pricing changes. Not the fact that everyone is getting charged 10x more for the same requests that they were using before. And a month pro subcription now runs out in 3 days, and a pro+ subscription runs out in 7 days - why even call them monthly subscriptions anymore?

I think I’ve made a final decision to move to Kiro.dev and never look back.

Sorry that isn’t what I asked for and I was trying to sincerely help.

They won’t acknowledge it, especially not publicly, as it can have legal repercussions.

if a user can spend 4-5k a month, the highest guaranteed plan being 200 is not enough

@banth31 Sorry but this thread is about individual requests and token counts. Could you create a separate post to discuss plans and tag me there?

image

My cache read is 60x input tokens. Isn’t cache read at like $0.30 / 1M though?