Doesn’t matter what version i use same issue,
| March 9, 2025 at 09:38:51 AM |
{toolCallComposer:{modelIntent:claude-3.7-sonnet-thinking}} |
Usage-based (392eabec) (but errored, so not charged for) |
| March 9, 2025 at 09:35:26 AM |
{toolCallComposer:{modelIntent:claude-3.7-sonnet-thinking}} |
Usage-based (392eabec) (but errored, so not charged for) |
| March 9, 2025 at 09:33:09 AM |
{toolCallComposer:{modelIntent:claude-3.7-sonnet-thinking}} |
Usage-based (392eabec) (but errored, so not charged for) |
generated for a second then stops. or Just sits there generating forever doing nothing…
like why does everything work amazing when i don’t need it to but the moment I’m actually trying to push trough my code and get some stuff wrapped up before i go into prod and… brick wall 2 days now unable to do anything in this project, doesn’t matter if i start a new chat roll back version nothing works…
and when it does it wont edit code in the file just keeps typing it out in chat. like why… i don’t want you to type out 1300 lines of code in chat when i asked you to fix one line.
when trying to fix lint errors it just completely ignored the errors themself go on a wild rampage, then it will fix the linter errors then says “Let me try once more with a complete replacement of the problematic section:” continue on a rampage ruining everything even more??
This is becoming super annoying very fast, like 3/4 of my time has been wasted by failed to generate, poor code application and context understanding in the past week, you explain the same thing 20 times 20 different ways with well laid out .md guides rules to read those .md guide direct rules on what to do, all is ignored, constantly saying yes i read that when it reads the first 50 lines and ignores the rest then you force it to read the full thing it says ahh i missed that part before then continues on with no memory of nothing, never ending pointing out the same thing i already pointed out was wrong with zero change, its like talking to a stubborn child… that refuses to listen but lies to your face saying they did.
I’m not asking it to do any complex tasks we are literally reading over php files and making sure security standards match across every file, they are all literally the exact same in every way other then the sql query’s and still cannot grasp the basics or follow a simple rule like.
“Never change the existing code unless it is to update it to the new security standards. we are not optimizing or refactoring, the code works we are just securing it. targeted edit only.”
end of rant
hopefully i can get back on track if things would just work properly…
I’m not quite ready to jump ship yet just a little annoyed, Love cursor its made me so much more productive and the amount of code i can bang out in a day is awesome “when its working” its just having a stable version that works 100% all the time would be ideal “and i do see they have started implementing these measures so that’s awesome!”
I know some of this issue is no fault of their own but Anthropic themselves, hopefully these problems will be addressed by both sides in the future, i know the demand is wild right these industry’s are exploding and frantically trying to catch up. Only time will tell i guess…
Bit annoyed that all of these failed calls have cost me so much money this month even though most say not charged for i know there is a great number that i have been charged for
“and other people have noticed this also”
, the fact that i have 1274 additional additional request by this time of the month is wild but i know its on me a bit to due to letting 3.7 go wild in composer sometimes.
“but hey when everything is working great and maintaining context it can be a absolute beast, I’m absolutely amazed some times when you refresh the page after telling it something that’s decently complex and see amazing styling, it implemented everything perfectly, made the db tables, got back end php and front end linked perfectly, and it works as intended, full security and best practices implemented, just blown away when it works like a charm”
Just gaged when i actually need it to work as I’m on a bit of a time crunch to get something wrapped up it just ■■■■■ the bed and i lose 2-3 days of productivity.
started working fine rest of the day today so idk… guess its just hit and miss right now… we will see…
1 Like
Have you tried with different model? I get this error regardless of the model if i try to use agent mode.
yeah doesn’t matter what model i use when its happening.
But i see a correlation to peak use times and Anthropic servers are just overloaded and cant handle the load, resulting in failed chats, failed edits.
or sometimes its like it became stupid af like they pushed you onto haiku or some other model that is just not up to par with what i need at that time.
so hopefully things stabilize soon, until then I’ve just been working on projects early in the mornings, mid day when it ■■■■■ the bed usually i just work on something less demanding, and late into the night when i see everything isn’t so overloaded. Weekends you cant even touch it during normal hours.
but its funny my main project that required large context will constantly fail but then i can be working in a different project that is just my front end code like 15 files and works perfectly the odd failed api call but for the most part works every time.
what models do you mostly use?
Interesting, there’s a known issue where having a large context will keep on generating forever and it may be paid too, do you try to let the agent work or give all the context at start? the other issues about Anthropic failing are well known too and Cursor can’t do anything about them, lets hope they build their infra on opensource models like R1, imagine if they could finetune with recent documentation or getting paid by companies to finetune on their codebase.
strictly 3.7 thinking now, odd times 3.5. But they both seem to fail just as often.
I prefer to use 3.7 as 3.5 just doesn’t work as well for my needs.
1 Like
I generally formulate a base plan let it go search for relevant files let it work to come up with a plan for what we are doing we build context only for relevant needs in that chat, i don’t give it access to the whole project scope as that’s way to large.
even doing it this was after a while context gets to large gets dumb fails, then i have to start a whole new chat “this is great if i was done with that section of code” but when I’m not and have to then make a new chat and try to get it to understand the current context by passing the composer to the new chat, never works then you get to painfully slog through the rest of the section like your monitoring a bad child.
1 Like
Guys ! Go back to 0.45.17 and you are happy !
0.46 is not yet working and has unlimited bugs…
Try a good prompt that summarizes your composer chat and skip Cursor summarization as it strips too much, also try with 0.45 as suggested upper to be sure its not a context bug from recent updates
I have been using 0.45 this entire time i switched to 0.46 for 1 day and instantly went back its not entirely linked to the cursor version. still experience same issues.
using 0.45 entire time only use 0.46 for 1 day it was terrible…
yeah i will give that a try, im starting to just have it make a good .md for each section of my project and placing it in that folder so i can just ask it to read it over, does seem to help “when it actually reads it in full”