Until a few days ago, I had the option to select mutiple models when running a prompt against an Agent. That option seems to have gone away. I am also seeing a lot less models when I go to the Add Models section (for e.g. I dont see GPT-5.4 High anymore).
Thanks for the question. You can still run the same task across multiple models, but there are two things going on here.
First, this feature depends on Cursor recognizing your workspace as a Git repo that supports worktrees. Please make sure:
you’ve opened Cursor at the root of the Git repo, not a subfolder
the repo has at least one commit
you’re not using a multi-root .code-workspace, since multi-model/worktree runs are not supported there
To make sure this is the case, check whether this drop down is clickable. If you can’t click into the dropdown, then you’re not in a recognized Git repo.
If Cursor is not in a recognized Git repo, the multiple-model workflow will not show up.
Second, on recent 2.7 nightly builds, the old multiple-model UI has been moved in the editor agent chat. Instead of the previous button/dropdown, you can now use:
/best-of-n<model1,model2,...> <task>
That runs the same prompt across multiple models in parallel. Try to get the model name as close to the name as possible, but if you just say, Opus for example, Cursor will figure out the canonical model name.
Regarding the GPT-5.4 High Disappearing, you just need to hover over GPT-5.4 Medium 1M and click Edit. From there, you can change the reasoning from Medium to High. Rather than appearing as a separate model altogether, the UI has been slightly updated to make switching reasoning abilities easier. So you still have access to the model; you just access it slightly differently.
Hi! A couple of weeks ago, the workspace option from the dropdown disappeared, even for old repositories that I worked on before (with multiple worktrees run). The command you mentioned (/best-of-n) is cool (I just tried it) and I will use it, but it solves a slightly different problem. Sometimes I want to see different runs of different models with the same prompt/plan and compare them manually and play individually. Is this feature gone forever?
A little update: in the new 3.0.9 version, the “worktree” option appears (and the “cloud” option disappears, weird), but it is disabled and recommended to use the /worktree command. Now I can’t figure out why I need to pay for Cursor IDE if it has a lack of UI controls and is moving to a CLI runner style interface, especially when there are better price/token CLIs in the market. I don’t need an ugly, text-bloated CLI; I need a nice UI. So, just IMHO, Cursor is losing its main difference and competitive advantage. It’s not valuable for me anymore =/
