Pricing Megathread and Q&A

[joefaron]

I’ve been away for last month, came back, only to find that I ran out of “credits” on pro subscription quickly.. just increased my spend limit, and then started to watch the cost go up.

i’m really bummed about current performance..

I spent about 3 hours on a master prompt, highly detailed, one to build an app on php, the other with python and flask.

the flask verison failed to install dependcies on windows, had to downgrade sqlalchemy, then downgrade a bunch of other things.. first $1.5 spent on it failing to install simple dependecies.. of which it normally would get on first shot.

the single shot prompt now for python/flask is at $3.45.. and its at this phase right now:

5 to do’s for $3.45 = $20.. for i guarantee a heavily broken app.

this seems not even close to what we’d get before for the cost and code balance.

i’m not sure if its you guys moving to this todo list, the cost of claude 4 sonnet thinking.. or what.. but i’m ready to leave.

i’ve already heavily struggled with coherence on any decently large app.. these costs are outrageous tho.. slow requests often made cursor worth it.. and now when its in auto, its really unclear if it’s using claude 3.5 and really how good it’ll be going back that far in time with ai models… as i’ve not used that in months.

anyways, for the first time i’m looking for alternatives… i’ve been a huge cursor fan for awhile now. i hope you guys can give me some credits back for the garbage code and wasted time..

if you’d like any further details or me making my requests public and giving you the id’s i can.. but i think you guys are in damage control and seem to be taking user feedback seriously. hope this is somewhat helpful. i’ll follow up in an hour seeing what it ended up with.

UPDATE 1:
charges stayed where posted…

1800 lines of broken php code and faulty tests..

2900 lines of broken python code… gave clear live examples of where each ones source code is hosted live for testing… scared of testing this further given these current results and costs.

[Cam_V]

Seems so. It was written on their website that local limits would be topped up every 5-24hrs, however this only seemed to have been in practice for a few days and the claim was then removed from their page with no explanation.

[fusiondev]

A month is like 10 years in AI development. 3.5 vs 4.0 is night and day difference, don’t even add his big brother opus maximus to the discussion. You need to be using the best models, with your to do list it’s massive, did you instruct it to create it? You should use Ctrl K n terminal to install the dependencies. What are you messing about with php for its not 2006.

[joefaron]

Hah.. it’s really the first time I’ve tried php yolo.. over 70% of the web is still rocking it brah. Plus my remote servers can easily test the code.. I was curious if I prompted it clearly, which would turn out better.. Turns out neither. I laid out a 600 line .md in plain English with app flow, general file structure, routes with ASCII template design and components, schema of existing sqlite3 db, along with plain English test cases. The idea is writing it clearly, and then following it up with saying what language to write and test in.. and comparing outcomes. Python and flask are using outdated versions, which is to be assumed sometimes with models.. whether it’s using an old js lib, etc… I did try one shotting it tho. Prompt was basically to follow TDD approach, break up the readme into phases and tasks.

Idk.. I’ve seen incredibly one shots first hand.. and things feel off right now.. not even considering the cost incurred for nothingness. Guess I ■■■■.

[QTom (Adrian)]

I have two accounts, but I received two different limit.

You’ve hit your usage limit

1. You’ve saved $95 on API model usage this month with Pro
2. You’ve saved $95 on API model usage this month with Pro.

[changxiaoqiang (ShuaiQChang)]

Kiro is so cool.

[joefaron]

competition is good but should try to keep this helpful / constructive critique for the cursor dev team. Kiro has tons of backing with amazon it seems, and can be bleeding trying to grab market share from cursor.. i assume cursor will do the right thing from here, as long as they retain top talent. <3

kiro literally reads your .cursorrules too hahaha.. so funny.. copying is the greatest form of flattery..

[leechoohyoung]

Recently, I don’t understand the basis for usage fee calculation.

Once a job starts, the usage fee is already charged.
However, if the job fails in the final stage (timeouts happen more often than expected), I ultimately don’t get the result I want.
In such cases, I believe the already charged fees should be refunded.

Is there any unreasonable point in my thinking?

[joefaron]

why’s it fail? do you timeout on running a cmd it gives? i normally dont see that..

[ammar26 (Hafiz Ammar Siddiqui)]

You should know your target audience is developers. Who can tell its a simple patched up wrapper over VS Code which you are overly charging now. Go back to simple pricing or get ready to be replaced by a better product soon

[liusuchao (Z)]

I hope you can clearly specify the pricing. Also, I don’t understand why you would do this. Why do you want us to leave?

[Lesslaw-m (Matt H)]

ok @danperks we’re 150 Q’s in, 6 days and not a single A

not very Mega

[jacky-wang]

I have the same situation. Can anyone explain for me? Is it still free or will it be billed in the coming months?

[Alxbrondj]

Finally, the whole answer to why the price has increased 20 times is in this topic))

it turns out that the larger the chat window, the more the cost increases. And here is a simple hint that will take up 150 tokens will already be 1 million tokens and 3 dollars. And so on, the next one is already more than a million, 1.5 million. Then just the question “change this color to this” will already cost 5$

I asked how to disable these cached tokens, but so far no answer)

Massive token increase on Ultra?

very interesting observation.

[Alxbrondj]

as they wrote, cached tokens are 10 times cheaper) Thus, the larger your context window, the more cached tokens you have. Even if a simple hint is 100 tokens, you will receive 1-3 million cached tokens each time. That is, you will write “Hello”, and pay 3 dollars :))
But I do not need cached tokens, how can I disable them? How can I return the cursor to a month ago? I worked fine a month ago and everything was fine with me. I was even fine with everything on March 1, 2025 If I roll back the cursor version to 1.0, will there be these cached tokens?

[DanEdens]

Even if you rolled back, they broke how the APi works entirely, There is no going back. Cursor is just dead. Since anthropic changed their rules, there’s no viable way to run cursor without paying hundreds and hundreds of dollars per day. They wont find a solution because there isn’t one. It’s over fam

[Abdul_Rahman]

these guys are doing some sneaky tactics if my suspicion is correct,

they are making cursor super laggy as your chat/context increases so you open a new chat with lower context,

they are inflating input output tokens,

they are using Claude 3.5 if 4 selected.

they secretly removed local limits, and auto is completely useless now, it was decent, it could edit 2-3 files in one go now it edits half the file and says it is done.

and now auto is also charging the user, it is consuming tokens and requests.

the change before this was perfect, I would get upto 600-700 usd usage out of my 20 usd subscription.

[nayabraheel]

I paid $60 for Pro+, and usage is showing 423 events. Now, Cursor is showing a message.

You’ve hit your usage limit

You’ve saved $160 on API model usage this month with Pro+. Switch to Auto for more requests or set a Spend Limit to continue with Sonnet. Your usage limits will reset when your monthly cycle ends on 8/8/2025.

What is Pro+ : ~675 Sonnet 4 requests ???

[QTom (Adrian)]

Currently, Cursor often wastes tokens on pointless summaries, while Claude Code completes tasks concisely and efficiently

[shrimplypibbles (Kelley Shayne Perry)]

i dont see a place to purchase pro+ only pro?