Cursor still great, but just got a lot more expensive

Well, it couldn’t last. I guess it was too good to be true. The first few months of using Cursor, being careful about the prompts I used, running out of the 500 included towards the middle of the month, eating the slow requests, paying for maybe another $30 of prompts. It would do everything I needed for $50, that was… great in many ways, because I was 10x as effective.

Then, the new pricing model : was just horrible. It basically translated to : as much gpt (which is so bad currently as to make 0.01x as effective) as you can suffer, and big bucks for Claude (which is where it’s at). I used my $20 and an additional $30 within 5 days of the start of the month, doing far less prompts than before, To get what I used to have as part of the pro plan (and a little supplemental billing) would change from around 50 bucks a month to more like 250. That’s a heck of a price rise for exactly the same product usage.

To be fair to Cursor, they did refund these unexpected expenses this time, and I really appreciate that : there aren’t so many fair and honest companies around these days who would do that. So, a big thank you.

But now, I’m still left with the situation, that to use Cursor usefully, as a real daily dev assistant, it’s going to cost 200-250 a month to get what I used to get for 50, now that even the legacy pro plan doesn’t include any slow requests.

That’s quite a lot of cash,

I’m going to look at a way to integrate the gemini cli into my workflow. That has amazingly generous daily limits built in. It will be a bit weird switching from Claude chat to cli commands, but it will keep my Ai code assistant costs down towards $50 a month. Which is about what I can afford.

It’s a shame that Cursor couldn’t continue to support the old model. Maybe the models will get cheaper and our bills will get less, maybe a competitor will offer a better deal, but for now this sees like the best I can do.

Has anyone else got any cost saving tips ?

12 Likes

TL;DR I need Ultra, but I’m not ready to pay for Ultra.


By the way, within just a week, I’ve nearly maxed out my Pro+ quota on my open-source and private projects too. Starting next week (or even today…), I’ll be relying on the mercy of renewable overlimit. But we just have to accept that guys realized their financial model was a bit too generous and decided not to bankrupt themselves.

You’re also welcome to check out my guide on cost optimization and make use of my Agent Tools to help streamline your workflow.

1 Like

34 million tokens :o ohmyohmy !

never tried going anywhere near that far, just assumed it would make a terrible job of anythign with that kind of context window. Thanks for sharing though, will check your repo. It feels like a lot could be done in efficiencies by using an mcp to create ready to use summaries for ingestion into future prompts.

So, after the refund, they charged my account again for the amount refunded. These guys really don’t know what they’re doing.

Vibe billing though innit.

10 Likes

I once tried to set up my mcp tool, but Agent couldn’t figure out how to connect it to itself, and I gave up on it.

Theoretically, I can try to implement my tools as a SaaS, but this can cause a lot of extra problems without bringing me extra money, so donationware the most convenient

I need refund, it’s not PRO, it’s TRIAL NOW!!

5 Likes

Have you tried Pro+ ? It’s $60 and I would say I fall into category of power user. It’s been about 10 days I think, I haven’t hit the rate limit. I use claude 4 sonnet, o3, gemini 2.5 pro. Sometimes max mode.

You can also choose Claude Code, at least we have a steady amount to use every day

8 Likes

Yes, I think it’s going to be Claude Code and Gemini CLI mashup route. That will give more requests, and be much cheaper. And mainly, I’ve just lost a lot of faith and trust in Cursor with this whole billing charade. They’re charging a lot more for what used to be cheaper. They maybe should have just been up front about that rather than try to dress it up as a new pricing model that was somehow “better” for the user. It’s basically unlimited gpt4.1 and paying 2-5x as much for the other models, the ones which are actually useful.

I’l consider a different plan once all the ■■■■ has subsided and they get a commercial director who knows what they’re doing.

1 Like

I see these clowns now removed the 24hour reset, I can’t wait for their apology message and how they didn’t convey the message properly. Please, don’t disappoint.

6 Likes

I’m using copilot, Claude code and if I’m desperate kilocode. Zero issues at all. Cursor has zero value proposition anymore.

6 Likes

They change rules now and then, these ■■■’s have no idea what they are doing.

1 Like

actually i just noticed i got charged for usage even though my plan is far from being exceeded and usage base pricing is off, wtf

It’s actually crazy how more expensive it got. I just used my 20$ usage in like 4 hours with Claude 4 Sonnet, my model I’ve used for the past months since it came out.
It just gave me 2$ on usage based. Before, this would’ve costed me maybe a few cents. Switching to Claude Code as we speak.

4 Likes

Lolz, let me know when you hit $800 in two weeks :joy: just turn on Opus 4

Honestly, i am confused. Are you flaming because you like their service but your just broke?

Cursor is a convenience, not a charity. There is even a free tier for starting so part timers can get something going and turn a profit and then reinvest that profit back into using cursor more…

This whole post screams skill issue and grubb.(even after them expensing your token usage as a loss… you still complained to stab them in the back…) Github is filled with free projects and open source tech, and your crying over having to pay for cloud tokens? … >_<

Bro host WizardCoder and open an API for cursor to use when you run out of tokens if your that broke… AI Coding is free… Quality deserves payment.

Wow, maybe English not your first language, but much like the AI, you’re failing to grasp first concepts here : I am upset that they refunded me, then billed me again for the amount they refunded, so the refund wasn’t a refund at all.

I think this processing language thing is a skills issue for you. Perhaps keep clear of it for a while.

I agree with you on your main point: Cursor is not a charity, it’s a business and a business has to make money. However: Calling people broke is a bit too far. Many people are struggling and Cursor was an excellent way to create something with very minimal finances. Relatively speaking it still is, one developer can build something now which usually took an entire team to build, so you can basically save $90.000+ on an end product given you’ve finished it within a year, so Cursor still hasn’t truly achieved it’s price-to-value. It can only go up. That means it’s slowly shifting from available to start-ups with no capital to a slight capital needed.

When we don’t project our situation on the rest of the world, it’s obvious that there’s many people from less fortunate parts around the world who might have great ideas but are limited in realizing it. Cursor provided the excellent possibility of realizing those ideas for far more people around the globe than it does now. Hopefully AI and thus Cursor will be cheaper in the future.

1 Like

On this we are agreed <3 cursor is still great. Nothing changed there.

Kiro also looking like good value, though yet to be proven. Looking at their pricing it seems to work out at about 47% that of cursor’s. It will be interesting to see if Aws can maintain that.

The pricing got very aggressive. I burned through my monthly budget in one session, grant it it was a about a 4 hour session.

And the fact that Cursor AI makes errors , or fixes one thing by creating a new error elsewhere, half the tokens go toward fixing the mistake it created.

And that barometer that shows your budget going down the drain is nothing but an anxiety inducer.

I enjoyed Cursor for many months but I haven’t used it once since these new updates took over.

I enjoyed my time with it but I’ve been looking elsewhere.