I have been using cursor for a long time and when I used version 0.44 and 0.45 I didn’t have any problems, on the contrary, I was happy to work with it. But from 0.46 to 0.47 I have been experiencing a lot of problems with cursor every day, today’s crash of claude api has knocked me out of my work schedule. it feels like no chat bot worked properly at all, except for gemini 2.0, even gemini 2.0-pro worked once in a while. The only working gemini 2.0 was giving such nonsense that I didn’t even torture it further to get something useful out of it. I decided to check how copilot was doing, and found that it had no problems accessing claude or gpt, and successfully closed the task in about 15 minutes, while I had been torturing the cursor for 2 hours. If it goes on like this, in all likelihood, I will have to go to the competitors, cursor team, it’s time for you to take your head, don’t forget that your competitors don’t sleep either, and your forum is flooded with a huge number of complaints and with each new version there are more and more critical problems, for example, the same problem with infinite generation has not disappeared, but rather on the contrary, only worsened
Yeah, the Github pilot also introduced the agent feature, and it seems promising. I’ve tried it out a few times.
Honestly, the copilot agent works pretty so-so, as well as the copilot codegeneration itself is far from perfect, you can see that it’s using pure chatbots without additional prompts, but at least it works, unlike cursor, and that scares me. Cursor is a really good product, but so far it’s flying into the abyss
It’s developed by Microsoft, and they have access to a vast amount of resources. Just imagine the product they could create after several iterations if Microsoft were willing to enhance the user experience.
It’s true that for so long microsoft has only just begun to approach its competitors, who are much younger than the same copilot)
This corporation has a lot of money, but they spend it in the wrong places.
Its now a disaster - nothing is working except for credit deductions
I had a similar experience today and cancelled my Cursor Pro subscription while opting for the full refund. I’m incredibly frustrated with the recent changes the team has implemented. Claude 3.7 seemed to be underperforming after the recent update, claiming it couldn’t even see my codebase and struggling with basic CSS tasks.
I switched to 3.7 max (didn’t even realise it was an option as updates seem to reach UK accounts later, at least in my experience) after seeing it was promoted as “smarter” in the IDE. There was no clear warning that max was usage-based ONLY. Since I already had usage-based pricing enabled (which typically only activates after exhausting fast requests), I didn’t receive any warning to enable it. The small popover that appears is not enough information. I assumed these would be covered by my fast request quota, just like the standard $0.04 requests are. I burned through 10x my usual spend in minutes, only realising the impact when I checked my account settings.
I feel for users who signed up for a year expecting consistent model performance, only to now be presented with extra costs to achieve the same performance they originally signed up for.
The lack of communication from the Cursor team is unacceptable. No clear/user-friendly announcements about the drastic pricing changes, outdated change logs, and a pricing history page that hasn’t been updated since March 17th, 2024 - over a year ago. I had to scour Reddit and forums to figure out what changed and why it felt like it was underperforming.
The performance difference is stark - the standard Claude 3.7 now struggles with basic tasks where it used to perform fine. Whether intentional or not, it feels severely downgraded compared to before. $0.05 per tool call AND premium request is excessive for max, especially when the requests seem to be consumed twice as fast as previously. Cursor claims the max model is only intended for “wealthy” users - to me this genuinely feels like a calculated move to force users to pay more to achieve the same blazing performance we’ve had over the last few months without the need for a “max” model. Perhaps profit margins were not where they wanted them to be, and this is their solution going forward.
Re VSCode above - VS Code Agent Mode now offers unlimited usage and tool calls for half the price, with what feels like a more capable, context-aware system. I’m moving to Microsoft’s solution. I’ve been a supporter of Cursor, but this update feels like a betrayal of their user base. I’d only consider returning if Cursor prioritised transparency about pricing changes and restored reasonable performance to the standard tier.
The new version of cursor that just went live in the last day or two is a major regression. I was using it the same way I have been for months and it deleted my production database. Cursor has never done that before.
I have it backed up so I can restore it, but it seems like whatever cursor did in the latest release the performance of the agent is totally screwed up.
Just to add another voice. I am severely disappointed, as I did pay for a full year less than 6 months ago.
It started with version 0.46, I used up all my fast requests trying to make something work. When I switched back to 0.45 it solved the problem immediately without any issues, but now I’m stuck on slow request for the past week.
Today I decided to try 0.47, and it’s still a disaster. I provided it with the files and it still could not read them, instead it gave assumptions. I tried to do a few things with it but it very quickly became obvious that it’s still just as bad as the previous version.
I switched back to 0.45 and will be on that version until it breaks or they release a new version that is at least as good. I won’t be holding my breath though, and when the time comes I will be thinking twice about renewing my subscription.
At first I disagreed with you, but now I agree. Yes, everything works much worse on version 0.46 and newer. It’s just stupid af…
how to disable update?
I just downgrade but it auto download new version
Me too… I am on my last threads… I used to use Cursor with joy, last fall, since about end of the year… it no longer allows you to choose the edits to make to files, it just makes them all, no matter the mode… that is supper nuts… I asked or a sniper rifle or perhaps semi automatic, but when I pull the trigger, it goes into full machine gun mode… spraying code and changing every file it can think of… sad, was really a great experience in the fall… thinking about trying windsurf, anyone try that?
I feel the same. If the support team does not try to handle it fast I don’t see a reason to continue using Cusros with all these frustrating updates/bugs.
lolll i will never not laugh at this sorry
cursor devs i am upset with you (ily but you are abusing me/my codebase/my copilot/my wallet) and frankly im going to post memes about it
why do i have to restart cursor every 3 prompts? are these failed claude calls still billing me? i think i spent $20 in compute in the last couple hours wtf are you guys smoking
Cursor is downgrading and aligning with Copilot. If Cursor doesn’t improve, Copilot at $10 with unlimited queries is the better choice.
Its weird, we have users complaining but personally for me, its supercharged my workflow even more, been one shotting features, not sure if its due to the cost, or just different skillsets/prompts/usecase for when to use modes on cursor that separate us
It’s terrible today it cant’ even follow any rules it’s gone backwards big time i think they are maybe giving me a different model to save on requests so i’m about to leave i have work to do, there are too many problems with it and not enough solutions it’s everyday they break something
Today it will not follow a single rule it is installing and deleteding packages for node and can’t actually code anything it’s all broken were as days ago a got a lot done.
this is a mess