Cursor is a team of young and smart people who formed
just a few years ago to build such an amazing tool
While this opaque pricing gives a poor user experience, and maybe it was a poor decision, everyone makes mistakes, learns from them, and grows to do better
Regardless of the new pricing model or the old pricing. For me. It’s long since paid for itself. I can’t really imagine how on earth this tool is profitable as it is so much cheaper compared to calling the API. In the picture I just opened a twenty dollar subscription.
I had made an AI chat type URL and then realized it was too costly. Big models wanting to have context require very large input and cache tokens, which cost more to input and cache than output, and I can’t believe that,Maybe that’s why so many tools are priced according to tokens!
dam, and i thought my ultra long 3 mil tokens rabbithole prompt was a bit over the top haha, 87 mil, holy f. what did you do? refactor the whole code base three times in one prompt?^^
it’s a valid way, user trust is important but they can’t keep not making changes, claude 3.5 to claude 4 model request once the cost has increased a lot.
I remember last year’s AI could only do one unit at a time, more than that and there were all sorts of errors and so on.
I don’t know, something feels off.
There are people who subscribed to the tool and on the same day it shows their plan covered $102 — that definitely shows there’s still a lack of transparency.
They’re a wrapper, so how does a $20 plan let you use $295 worth of API?
Does that make sense?
At what point are they making money — or are we the ones losing?
Calling the API way to use $100 a day isn’t much, any tool I’ve tried it is, and in less than an hour $10 is gone (sonnet)
In terms of API pricing. I think we still made a profit.
If you charge at twenty dollars for the API, you run out of credit in a few hours, and anyone who has used their own API knows that’s a horrendous cost
This is an inevitable result, no tool can subsidize users forever, it is likely that this year or next year other IDEs will also gradually increase prices
Last year to now, as the AI modeling capabilities improve, a conversation can perform more tasks. Means consuming more tokens API cost remains the same, but tokens skyrocketed and the cost doubled several times.
In fact, $20 can buy 100 conversations is good, now a conversation consumes tens of thousands of tokens on average, more if you need to search the web…
Here’s the thing about the tokens. Cursor adds layers to this that inflate the tokens by 10x. Go use claude code and watch your token usage plummet.
I get a lot out of anthropic subscription now, kind of like how cursor used to be. Now cursor isn’t even worth the 60. Id pay 60 if it was worth it but 20 at Anthropic does more. Cursor not showing UI of when a person can work again is perhaps the most ridiculous thing.
Mod Edit - Isaac - Please don’t spread misinformation. You’ve most likely misunderstood Anthropic’s inference cost pricing difference for their own models VS Cursor having to purchase these at an increased price from what it costs the model provider. This does not equate to cursor “inflating tokens”.
I think there is a huge disconnect in how people think this works. API pricing is NOT what they put. That’s api pricing for individuals without business emails. Businesses with large spend get MASSIVE savings on compute. Up to 90% cheaper for those spending millions.
Look at Claude code. Just spending $200 gets the user 20x what they spend $20 for. That’s 50% less money for the compute. These companies are buying millions of $$$. They are not charged what we are charged. They are making a lot of money.
Mod Edit - Isaac - Again, you’re using an example of a model provider comparing their own inference costs, which could also be at a loss as far as you know.
I think the cursor team should be more cautious and transparent about pricing strategies, rather than repeatedly modifying and hiding detailed terms, and simply listing a few sentences on the pricing page.
The reason why paying users are willing to pay is that they accept various membership level value-added information at the time of payment, which cannot be easily changed or even made worse!
I guarantee that no paying user will be happy with this very unprofessional behavior!
I understand that cursor urgently needs to prove its profitability to investors, but this is not like your software version iteration where a new version is released every few days. Business strategy must be very cautious because it involves money!
Yes, a balance needs to be found and transparently disclosed, billing by toknes will ensure long term profitability.
Billing by conversation count, I think that will slowly all go away over time, the model’s ability to come up. A conversation can do more content and consume more tokens. the average cost of a conversation skyrockets! The claude code is the cheapest, after all they are sold at cost.
I think I’ll opt out of the new pricing model because it just doesn’t make sense for me it’s honestly too much. I tested the system by simply sending “Hi” and receiving a small reply, and it cost $0.22. The team really needs to review this system seriously.
I’m still confused. Am i to understand if i opt out the new ‘payment system’ i will still get the 500 requests (as long as i have a model where 1 reply from me = 1 token, like sonnet 4) as fast requests, and then unlimited slow requests?
Yes to this. The dashboard that Cursor put up are not their actual costs, there is absolutely no scenario in the world that would be exposed to the users. imo, it was better not put up these numbers at all, as this feels like a “fake” way of calming down the users — look at all the savings we are giving you. And again, that is just obscuring the truth.
I truly wish Cursor success, and hope the team finds a way to come back to this. Personally for me it would be enough to just have more clarity on what the limits are, when those are reset, etc.
In my opinion having 5hr blocks of usage limits (like Claude) is a better experience then monthly limits. It removes the possibility of blowing your entire stack on one bug.
Yo mod. It’s not misinformation. As a business with api pricing, the costs plummet per token AND the token per prompt lowers with claude code Cursor does NOT spend consumer api pricing either. Anyone who continues a project in claude code will see there token usage PLUMMET vs Cursor. They will see the problem is the Cursor wrapper inflation of usage. YOU don’t over moderate and stop others from knowing the truth.
Again. You are comparing base consumer api pricing and not business pricing. Get the high usage business api prices and use those. THAT is what cursor pays. All business that use high tokens get huge compute discounts.
This is exactly what happens when a company scales rapidly without a profitable business model from the start.
I have been a Cursor Pro user for 10 months, since the beginning, and I was very happy with it. However, around the time Claude 4 was released and they began introducing features to become profitable, such as the MAX mode, things started to deteriorate. The agents became less effective, forgetting information and losing context. I suspect this was due to Cursor trying to save tokens by compacting the context.
I then decided to try the Claude Code $100/month plan, and it has been a game changer. Although Cursor also allows the use of Sonnet 4, I now realize that context is crucial, and CC excels at maintaining it. Additionally, I appreciate not having to choose a model, which can feel burdensome with so many options. I prefer to keep things simple, and CC does a great job of that.
I am not sure how effective the Claude $20/month plan is since I haven’t tried it, but if you can afford the $100/month option, I highly recommend it.
In the end, Cursor is just a VS Code fork with an API to call third-party AI providers. I believe a company like Anthropic, which develops its own AI models, can offer a superior experience since it is not constrained by third-party costs, only its own.