Hey, not sure but I am experiencing a lot of claude not being able to edit things and saying “no changes made” especially with files more than a 1K lines. I wanted to see if anybody else is experiencing this.
1 Like
Same here. Often times the loading indicator at the top left of an edit box never goes away, almost like the model tries to edit a file before the last edit took place.
There were adjustments at launch and some heavy load occurs, is it still happening?
yeah, it still happens. it’s annoying because the only fix is to revert the whole request and start it over with fingers crossed.
Could you post one request ID with privacy mode disabled temporarily so the Cursor team can have a look what goes wrong?
This is deeper technical issue that may need to be analyzed. Since it doesnt accur for everyone its hard for Cursor team to see what went wrong without a direct example.
what is “privacy mode disabled?” I have the request ID, but not sure where that option is
nvm found it.
6956bfb4-0066-4864-b5f9-85907c1136aa
here’s one from after I disabled it:
157c7e69-085f-47a0-824e-bf819041b75e
here’s another one:
4015605b-8b7b-4f77-9d61-34940de89213
and another one:
5bdca46b-4841-4463-9627-d273085b23fc
here’s the most straightforward one yet:
97096fad-7bd9-437c-ab3c-e0bb2dc4f570
here it tried to revert a file with git checkout because of the editing errors. it would be nice if we could revert up to a specific code block; just rejecting edits confuses the request context.
bb7a30c1-342b-4055-9f06-bb54e2bfd93d
here it failed on two different files + connection issue on resume. starting a new request does not solve the problem.
f7af900b-37cb-49a7-a4ff-acbcb5571b45
1 Like
@deanrie could someone from the team have a look at the request ID and check what happened with edit tool?
cc @danperks
Hey, this should be improved now, sorry about this!
There may be a need for an Model EAP for those who want to access the latest model even its just fresh released while everyone needs to do proper testing to find issues and adjust tooling. 
Others who are happy to use well tested models could go the standard route and get new models once they are tested and tooling adjusted.
Hey, it’s been working pretty well since you posted this, but now I’m getting the same issue again:
6bfa9d76-7ad9-4b17-bec2-95d7278aa849
in fact the problem seems to be model agnostic. here’s gemini 2.5 pro:
6174e561-faec-41f2-85ca-6ef3e3da0c2b
Hey, it looks like these errors are actually somewhat “expected” as the AI will sometimes call the tool with the incorrect parameters, which means the tool call will fail.
It’s a hard one to fix fully, as there’s always a chance the models like to hallucinate and get the output format wrong, but I’ll pass this to the team to try to improve on!
“We’re having trouble connecting to the model provider. This might be temporary - please try again in a moment.”
Help ! 
fix claude 4
It feels like claude 4 suddenly became very silly on the cursor today.
2 Likes
bumping since I am still noticing the issue 05/28 pretty frequently. thanks for your help so far!
Dan mentioned yesterday in another thread that they are looking into improvements for this case.