I just started with this tool and at first everything was really good using composer 2.5.
I had a popup asking me if everything was fine , i answered yes.
Not sure its related but now i cannot even use this tool anymore. Its gone stupid, remains me the SWE 1.6 from other or the copilot. I have to verify all and always problems, mistakes, big errors, and thats when he understand what i ask.
Again its was working perfectly fine when i started. i change chat every time, so its fresh, use git.
It looks like every month i have to search for the new tool but cannot continue with this one. Anyone with the same problems recently?
Hey, let’s clear this up. First, about the popup, it’s an in-product satisfaction survey, just telemetry. Your answer yes doesn’t change the model, routing, or quality in any way, so it’s definitely not related to any degradation.
About the model itself, on the server side everything looks clean for you over the last week. I don’t see any errors, and usage has been stable through Composer 2.5. What you’re noticing is usually tied to the context of a specific session or the task complexity, not a model swap.
To dig deeper, I’ll need a specific example:
The prompt where the model did poorly, and what exactly went wrong
The Request ID for that response right top corner of the chat > Copy Request ID
Your Cursor version and OS
A couple things that often help:
Make sure the chat has the right context, add the relevant files via @
For unfamiliar or large tasks, try Plan mode. It plans first, then executes, which noticeably reduces mistakes.
Send over an example with the Request ID and I’ll take a look at what’s going on.
I was using copilot for almost a year with no problems and suddenly in may it started to decrease in quality, then the price changed and was even worst. not only the quality down but the price up. So, I switched to Cursor, it started pretty good for a month and then, the same exact thing, the quality decreased drasticly.
One day, I can ask complicated things and works perfectly well, the next day is impossible to ask anything.
when that happens i try to switch from composer to gpt5.5 or claude 4.6, but does the same… nothing.
I try to open another chat and start over, got the same.
If by miracle you have good chat with some clever AI behind, you are back again in business, you can ask anything and works but thats rare now.
I get it, instability is really annoying, especially when everything was flying yesterday and today it’s dragging.
One important point. You said that on “bad” days switching to GPT-5.5 or Claude 4.6 gives the same result. If all models drop at once, it’s usually not the model itself, it’s the session context or how the task is phrased. On our side, your account looks clean over the last week. No errors, stable usage through Composer 2.5.
To actually find something, I need a concrete example right when “nothing works”:
The Request ID for that bad answer (top right corner of the chat > Copy Request ID)
The exact prompt and what went wrong
Your Cursor version and OS
Without a Request ID I’m basically looking at nothing. With it, we can check what happened for that specific request.
A couple things that often help with stability:
Add relevant files to context via @, don’t rely on the model to pull everything in by itself
For big or unfamiliar tasks, try Plan mode. Plan first, then execution, it usually means fewer mistakes
Send an example with the Request ID during one of those bad moments, and we’ll check what’s actually going on.
When model quality feels like it suddenly dropped, I try to separate three things before blaming the model: context, task shape, and verification.
A fresh chat helps, but it does not guarantee the right files are loaded or that the task is scoped well. For a bad run, the most useful evidence is usually one concrete prompt, the files it saw, what it changed, and the exact mistake it made.