Explore agents ignoring model variant

Where does the bug appear (feature/product)?

Cursor IDE

Describe the Bug

Only during exploration, even though I have my subagent and explore agent model set to composer 2.5 (not fast), it always default to 2.5 fast. Seeing as the fast variant takes ~3x more usage than the normal variant, for us on the pro/pro+ plans it can make a big difference.

Steps to Reproduce

Set the explore agent in Settings → Agents → Explore subagent model to Composer 2.5, hover and click edit, disable Fast.

Ask for any task where exploration is required, and you will see the agent is running as Composer 2.5 Fast.

Expected Behavior

The agents should be spawned as Composer 2.5 without fast mode enabled.

Operating System

MacOS

Version Information

Version: 3.5.33 (Universal)
VSCode Version: 1.105.1
Commit: aac81804b986d739acab348ed96b8bea6e83cc50
Date: 2026-05-22T06:47:48.039Z
Layout: glass
Build Type: Stable
Release Track: Default
Electron: 39.8.1
Chromium: 142.0.7444.265
Node.js: 22.22.1
V8: 14.2.231.22-electron.0
OS: Darwin arm64 25.5.0

Does this stop you from using Cursor

No - Cursor works, but with this issue

Hey, thanks for the report and the clear steps. This is a known regression. The explore subagent set via the UI with Fast turned off still resolves to Composer 2.5 Fast. I can reproduce it, and we’re tracking it on our side, but I don’t have an ETA yet.

Related threads for context:

Once there’s an update on the fix, I’ll post it here.

same bug being observed in 3.6.31.

This is a bad one as it actually racks up a lot of extra costs for long-horizon tasks where subagents keep getting spawned. The only solution was to switch to a non-fast model in the settings

Thanks, confirmed. This is the same regression described above, and it still reproduces on 3.6.31. We’ve reported it internally and are tracking it, but there’s no exact ETA for a fix yet.

On costs, I get it. In long tasks where subagents keep spawning, the Fast variant can really add up. As a temporary workaround, you’ve already found a good path: pick a model that doesn’t switch to Fast. I’ll post an update in this thread as soon as we have one.