Ever since X took over, the auto mode has become really sluggish, really slow.
Browser usage has plummeted.
I keep having to ask my chats to proceed/continue, because they end up as “reading…”
Is there any big reason? Is it because “grok” replaced Composer 2.5 using auto most of the time?
Hi @master3395 Thanks for the report and sorry for the hassle here. I did some research on your account and usage and found some interesting things that might help you.
One cause of slower responses on Auto is messages that include images: screenshots you paste in, and screenshots the agent takes when it uses the browser. Auto and Composer 2.5 often run on models that don’t read images directly, so Cursor first turns each new image into a text description before the model starts responding. That step can add several seconds before you see anything, and if it happens partway through a task it can look like the agent is stuck on “Reading…”. If you often work with screenshots or the browser tool, that’s likely what you’re seeing.
A couple of things that should help:
- For screenshot-heavy or browser-heavy work, pick Grok 4.7 or Grok 4.6 in the model picker instead of Auto. Both read images directly, so they skip that extra step, and both are in the Cursor Models usage pool. Composer 2.5 won’t help here because it doesn’t read images directly either.
- When you can, paste error messages and logs as text rather than screenshots.
If a chat still stalls after switching, could you grab the request ID (open the chat’s “…” menu and select Copy Request ID) and post it here? More on what’s useful to include: Reporting a bug | Cursor Docs
Also, could you say a bit more about what you mean by browser usage dropping? Is the agent using the browser tool less than it used to, or you’re using it less because of the slowness experienced?
Thanks for the detailed feedback.
I am still trying to use the browser tool, just like before.
But now it’s either not responding to use the browser tool at at all, like it’s working without showing it in the browser, that’s fine, but it looks stuck, like i asked it to use the browser to make a pr on github, and it never took charge on the browser.
And on another chat i asked it to make a token, to update a docker i have.
I ended up creating the token myself, and pasting in a screenshot of the token, in order for cursor to read that, and continue with the work.
It was in this chat:
Request ID: a83399d3-6c10-4b50-905b-6762c3d3b90b
I see now that auto uses sonnet 5.5 it is much faster.
How come using MultiTask does not work as well as it did before?
Is it really slow and/or getting stuck using “auto”?
So it piles up, right now it’s at 8.
And my Virtual Box crashed, because it wants to “Think” for several minutes, and or never continue?
And since i run “multitask” i can’t just post “Proceed with fix WordPress scan missing wp-test” or something similliar?
