Glad to see I’m not alone in my frustrations with 0.46
Initially I thought “oh, this new version is much smarter”, until I realised it was just pretending to be!
Not being able to add a specific folder @path/to/folder to the context is a miss for me, but I would accept it if I felt there were enough improvements to compensate. Alas that is not the case.
0.46 must also be much more resource intensive because it freezes frequently.
As many others, i’m rolling back to 0.45. 0.46 was a terrible experience.
The return to 0.45 helps just as little. Code quality, context, etc. will not improve just by resetting the version.
The context performance is insanely bad. It looks for files that are unnecessary, it even stucks in a loop sometimes doing same things over and over again…
I’m truly in love with Cursor, but after two days of the new version, I lost the composer goin into loops of installing and uninstalled 0.46 version three times until I arrived at the forum and saw that many other folks are experiencing the same problem as me. After rolling back composer appears again.
I’m very pleased with the workaround, but please Cursor Team get hands-on it. We, the Vibe Coders, need to move fast…
Have a great afternoon, the issue just took one hour to fix. Thanks again!
Thank you so much!
I am experiencing frequent crashes, particularly with the cursor, which hinders productivity. It appears that updates are being pushed without sufficient testing or logical improvements.
Could you please address these concerns or provide an option to revert to a more stable version?
Looking forward to your response.
Hey there!
I’m in the same boat. I wasted half of my pro subscription in just two days because of chat hallucinations and useless requests. When it works, it’s great, but the anthropic rate is way too high!
I’m planning to downgrade, but I’ll only use it in slow mode.
Are there any ways to get a quick refund for my request? Haha.
I’m in the same situation and to be honest I regret having paid one year up front for cursor. The 500 fast requests (or in case of sonnet 3.7 the possibility to call this llm at all) have been consumed this month in an instant due to regular crashes (error 5, etc. => due to long chats as it seems), cursor just stopping outputs in the middle of doing something without any idea why, regular failures in tool calls, regular failures in using commands (despite having used the commands over and over again correctly beforehand), anthropic not being reachable, and many more.
I have no clue if this is directly related to 0.46 but overall the experience feels worse. I also don’t like that everything is in one chat window (agent, chat, etc.) and not seeing your fast responses consumed somewhere next to the chat window is also really annoying. In the previous versions you were able to copy your prompt and if you re-used it, all the @ references you made were still there. With the newer versions, you have to be lucky that you can even paste the copied text into the chat windows and if it works it loses all @ references. So annoying!
I can only hope there is an improved version soon otherwise I pretty much paid in advance for a quite buggy product. Fingers crossed
Oh good to know it is a problem in the version. i began using those days and was even considering signing, but yesterday i began having those problems you mentioned where it began to destroy my app whithout me asking for it.
So i was giving up all together.
That is good that it was the version, so i ccan consider paying for it again if it is fixed. thanks for the info! ill dowgrade it.
This is the worst update i have seen on any product ever. I mean chat/composer was working in a perfect way with giving us ability to give the context as we please. Right now agent seems to be misplacing everything
I moved back to 0.45 and could breathe again 
Although, I do think I got carried away with the hype with claude 3.7 and latest version of cursor. Whereas before with 0.45, I’d take my time, plan and document the roadmap - all done by the AI of course in full yolo mode. But properly, as if I was working with my rl devs.
But with the latest version and claude 3.7, i’d be like “Make me an app that’s gonna make me 100 mil” and then get annoyed when it can’t even construct a decent looking UI.
But now im back to 4.5, everything is calm again and expectations have been reset. TBH, it’s still 100x the workflow of working with rl devs so i’m not complaining too loudly 
I’m looking here: GitHub - oslook/cursor-ai-downloads: All Cursor AI's official download links for both the latest and older versions, making it easy for you to update, downgrade, and choose any version. 🚀 which exact version of 0.45 is everyone using? 0.45.14?
I’m beginner.I’m not sensitive to the tiny difference .
But I love Cursor.I hope I can do good as you
completely agree. just rolled back to previous version
Tab stopped working for me, I tried a long bunch of things from the Forum and logged the issue. Only solution - rolled back to 0.45 and it’s working again. Shall not be updating for a long time.
I think Cursor should fork a version from 0.45 to fix bugs and release it as LTS, only fixing bugs and not breaking the core features.
I am back on 0.45 for a few days after major struggle with 0.46.
I was flying with my projects, then everything went to ■■■■.
Then came across this thread, thx god.
0.45 + o3mini now rules !!
Used to use claud3.5, 3.7 did not convince me. Did ambitious/unnecesar constructs, usually failed 
.
o3 is smart claud 3.5 for me :).
After burning 100 credits in 2 days with Cursor 0.47 + Claude 3.7, I have switched to using Claude 3.7 but with Cursor 0.45 and experimenting with another program whose name I can’t say on this forum but it rhymes with “binned-turf” and it’s air and water themed. The experience with Cursor 0.45 and the other program is slower and less flashy, but I’m able to incrementally add or change one thing at a time without the agent breaking two other things that were previously working. Claude 3.7 thinking seems to still cost 1 credit per use on Cursor 0.45, and it costs 1.25 credits per use in the other program. And in both programs I’m having fewer “context” problems than I had with 0.46 and 0.47.
As for the general complaints about Cursor and the cost, I am still gladly paying my $20/month. Even if it acts like a junior developer sometimes, $20/month is dirt cheap for a junior developer. And I’m happily paying the cost of the aforementioned Cursor competitor. It’s healthy to have competition. Progress isn’t linear, and sometimes it goes in the wrong direction as I think it did between 0.45 and 0.46. But as long as the developers are working to fix the issues and trying to stay ahead of the competition, as I think they are, I’m happy to keep paying.