Can we, for the love of God, get proper Jupyter Notebook support? The tickets have been open for nearly a year now, and it’s getting worse with each update.
2 Likes
I see in the example that the full-text editing in the video can select the model because I see a drop-down selection next to it, but why can’t my 0.50 version choose the model myself?
1 Like
Is there a way to bring back the old way of review changes? I’m not sure if it is because this update is full of bugs or what, but I can’t quite understand how this new way is better, no easy way to reject changes, files changes are piled one after another so it becomes an infine scroll, button to reject or accept all changes dissapears for no reason, etc.
3 Likes
While using Cursor AI with the Max model at version 3.7, I found it to be extremely inefficient. I tested it and burned through 500 tokens in just two hour, and it could barely handle a 6000-line JavaScript file. Honestly, these Max model pricing plans need to be adjusted urgently. It doesn’t deliver value for the money.
1 Like
Yep, I let it run unfortunately and it cost me $200USD of o3 max calls - right after the pricing change. I’d done equivalent work with o3 before and it cost less than $50… still too pricy now… as it was the only model not creating errors. Gemini couldn’t MCP call, Sonnet didn’t have the reasoning or awareness of the full project…
Cursor got greedy, at the wrong time too. Codex Agent destroys any output by Cursor, and OpenAI just acquired Windsurf; so, unless Cursor buddy up with Google or Anthropic at least, they are at a massive disadvantage.
There is so much bloat in Cursor now too. I can’t imagine anyone new to coding using it over the other choices (even Super Agents like Abacus.ai Deep Agent, Scout.new, GenSpark etc).
Free - Cline and Roo Code also do a better job of automation and understanding what you actually want. ironically using o3 with them costs me far less than the paid plan (oh yeah, not only is it $0.50AUD per call, it takes up your advanced requests too).
2 Likes
true
1 Like
I’ve been on Cursor Pro for about a month now, and overall it’s been fantastic. After my initial trial, I did run out of quota pretty quickly, so I switched to a mix of “slow requests” and the usage-based plan. For the most part, that setup worked great—very smooth and reliable.
But after updating to version 0.50, things got tricky. When I’m on “slow requests,” responses now take nearly a minute (or more!) to come back. And often I have to repeat my question two or three times just to get the code tweaks I need. It’s been a bit frustrating, since before the update I could fly through edits without any hiccups.
Has anyone else run into this? Any tips or workarounds would be much appreciated!
3 Likes
Personally I saw the opposite in performance, most requests are so much faster on v0.50 but this also has to do with the AI providers availability and congestion.
Note that the more often you use slow requests the further back in queue you land to make it fair for all users.
Where are the details in the changelog about the latest 0.50.5??
1 Like
Where is the in-app changelog?
So there is an issue with multi-root projects where the agent is unable to understand what file path it needs to use ie if the top level project is P1 but the task is in P3 it will update the files in P1. This issue is pretty severe within the Laravel framework and the agent ignores the file path rules/logic checks when aded to context.
Im sorry, but this is honestly the most frustrating update to date. It feels like the update made the models less intelligent and more prone to doing their own thing instead of listening to the input given. I’ve been using Cursor for about six months now and loving it, but this new update is making every task a chore and very frustrating. It introduces things that it was never asked to do, and it seems to error out mid request so much more than it did previously. Also, I’m utterly confused by the new pricing model. I’m having to watch my account minute by minute to make sure that I’m not being charged for things I’m not expecting. It seems like this update was a money-hungry cash grab that aims to end up charging the user far more than any other update. It’s very frustrating.
4 Likes
Hi,
which version is your cursor?
was removing Claude 3.7 slow requests part of the update?
Version: 0.50.5
VSCode Version: 1.96.2
I found the issue then. Basically, there was this empty model in the settings, and it was checked. I removed it from the list, and it’s now gone. I still don’t know what model was calling, though.
1 Like
I have seen that too. Thanks for the info. 
I hate to say it but… He’s right…
I now use Windsurf after complaining in this forum that Cursor wasn’t catching up with them. I will probably still pay for Cursor here and there as it is better in some regards, but for coding with an agent, Windsurf seems to be a lot better.
I’ll miss my unlimited slow tokens, but they run free agents and ■■■■ near free promos all the time. claude 3.7 is cheaper over there too and with your 500 requests allotted a month right now you get some crazy 4000 GPT 4.1 requests.
They also don’t have an AI API which is ■■■, but you know, Cursor is not catching up much at the moment.
I’m lucky I usually stay under $300-400 a mo on these tools atp.
can you set up postgres database for the background agent to use?
what you guys think abotu mcp server that will allow cursor chat agent to create background agents on go?
Just testing now, may opensource it for everyone
1 Like