I am totally a bot bro. Cursor drowns me in $$$ to share reasonable takes and try to balance out all the haters. But seriously, I pay $16/mo, and I get more than $16/mo of value. It’s that simple for me. Maybe those dropping hundreds a month on cursor are disappointed. But I am offering a genuine opinion as a solo dev who has tried other tools and uses Cursor 10+ hours a day.
Also I have given suggestions for other services (Windsurf, trae, supermaven) and asked for alternatives many, many times. Biggest tell that I am not some cursor shill. It is possible for some users to be mostly satisfied with Cursor, but still have some reservations.
But whole reason why I switched this month from Copilot enterprise to Cursor is that Copilot will CUT your files, when there are enough lines. For example 1000 lines file might get cut to 500, leaving it corrupted.
And don’t even dream on optimizing them by hand. These AI tools are so powerful now, that they do all coding and there is simply no time going through all their doings manually. In two days Cursor have produced over 47000 lines of text (12000 lines of executable code) and it sure ain’t very well distributed to different files. Copilot would not be able to handle this project anymore even if I just merely opened it there.
i think setting the agent to auto just uses a cheap or free model seems to work with the 1 month trial and i dont think its gona charge me any extra not sure though
With a paid subscription, Auto is free (but that ends 9/15 unless you already have an annual subscription), but I have also noticed that gpt-5-mini is free. So I use that one most of the time instead of Auto.
you can still switch to the old pricing. the option is gone, but instead you now need to write an email to them (if I remember correctly its [email protected]) and they change it for you.
I thought so too, since that’s the obvious and fair thing to do. I paid for the plan that included unlimited requests plus 500 fast requests, so of course Cursor should deliver what they promised and what I’ve already paid for.
But in my case it has not worked like that. In the thread here: Locked into downgraded plan - annual subscriber ignored by support, they asked me for my email, and then nothing happened. I followed up, and @deanrie eventually replied: “They replied to me: Unfortunately, we won’t be able to revert them back to the old pricing.”
Before all that, I had already sent emails to both [email protected] and [email protected]. I received nothing but automated responses. No one from the team actually replied to what I wrote, even though I clearly asked for help or at least some alternative.
I haven’t fully decided how to react yet and need to check with finance, but I am seriously considering disputing the charges. This is simply not the plan I paid for. I even mentioned in my emails that I was open to other solutions, like a pro-rata refund or something similar. Still, nothing.
So here I am, having paid for one year of unlimited agent requests and 500 fast ones, and now the plan has been changed entirely. Cursor says they won’t revert it, and they also won’t let me out of the contract.
To me, that feels like fraud and possibly criminal behavior. I never expected such openly customer-hostile treatment from a company I trusted. I’m honestly shocked.
Try making a post in this thread and be sure to give us an update. Being able to get a refund after they make a change to the previously agreed upon benefits of the subscription is key. Otherwise I may be in this situation too after the next change. Maybe if your charges are beyond the disputing window they just ignore you.
The cursor has become very expensive. It completely spends the budget in two days, with a very modest volume of requests and auto usage. And we often pay for the fact that the cursor makes mistakes and tries to solve them without success. I understand that developers need to earn money (I think developers have always earned money), but if no one uses the cursor, will you be able to earn anything?
Can you explain what are the nature of your requests (vibe coding vs assistant)? How big your project is? Are you giving it clear context? What models you are using? Do you ever try to use Auto or other free/inexpensive models like GPT-5-mini?
I don’t know your situation, that is why I am asking, but a lot of this could be a misunderstanding of Cursor as a tool.
You are definitely right, no way are you getting 500 GPT-5 requests for $20 of usage. They have done a good job making gpt-5 use less tokens, so idk maybe its closer to that number. I think most my gpt-5 requests use a minimum of $.07, so that is only 286 requests, and that is on much lighter stuff.
It looks like the cursor is aiming for a big fat enterprise rather than an army of independent developers. Cutting out the slow requests is a sign. I would expect that it will be even more expensive in the future. Honestly, the “Common people” of “Black Mirror” lightens it up very clearly. I would recommend that you watch it if you haven’t yet. This is what the Cursor really does in real life. If you are an independent developer and using the Cursor for your pet projects, I would recommend switching to your model provider. Cause the strongest thing in Cursor was a selection of providers for a reasonable price, which is not a thing anymore.
Yes, it costs, you won’t do any serious work with 20$ a month anywhere now, tokens are getting cheaper, but solvable features are getting more complex.
No, it is super cheap, the same human coder performing the same work would cost in my case 40x-50x more.
We can have this conversation when the base subscriptions are 2000$/month at least, because at that level we are getting human level salaries but robotic work which we need to scrutinize if it’s equivalent or not (it is actually better right now).
The vibe coders are the ones that are getting take advantage of. I have no need for vibe coding because my design/test/improve iteration can only be done so fast when done by a human, and AI makes too many errors or unmaintainable code to have full faith in not reviewing code, so that slows down the ability to “vibe code”. Some projects can be vibe coded, while others as of now can’t. $20 has been fine for me, but if I was vibe coding, yea that price tag is going to go up and up, even if it is cheaper than hiring a programmer, I think a lot of people are expecting their vibe coding costs to stay low but they aren’t going to. This has been a strategy by these AI companies to take a loss early on to get people to adopt and become dependent on their services.
If there were no show-stoppers, for you or for a vibe coder, in the form of code, integration and execution review, you could run the LLM 24/7, with multiple instances and use millions of dollars a year. In that case, it would cost more than hiring, but it would need to run 24/7 and not make too many mistakes.
Currently a large portion of my budget is going into automated in-task assisted linting, compiling, debugging, documenting and deploying.
For example a task can be one shotted with 4 cents. But then it will compile, build, test, review results, bugfix, self-document, report and git push a version increment all in a single chat, and run up to 140 cents.
So every time I vibe ask a small feature that is simple and nice to have, another dollar or two is added to the bill. And then move to the next vibe task. The faster the tasks are completed, the faster you move further, ideas never end, feature creep is infinite, LLM tokens will be sizzling, the limit is not your imagination now, just your wallet.
You bring up a great point. Vibe coding can often be a hinderance because you get lazy and decide to just add every feature instead of designing a clean and clear product, which is required when you have limited resources. The difference is like having a digital camera vs a film camera, pros and cons to each, but one made people become masters at the art. People can get a false sense they are developing a skill when in reality its just the AI tool flexing.
I thought Cursor was expensive until yesterday——I use Gemini CLI (Gemini 2.5 Pro) to generate code to seek answers like usual for ONLY ONE DAY and finally I found it cost me $17.
Now I have huge respect to Cursor as a Pro subscription user.
Cursor drowns me in $$$ to share reasonable takes and try to balance out all the haters. But seriously, I pay $16/mo, and I get more than $16/mo of value. It’s that simple for me. Maybe those dropping hundreds a month on cursor are disappointed. But I am offering a genuine opinion as a solo dev who has tried other tools and uses Cursor 10+ hours a day.