Auto-Balance chosen for sub-agent work instead of Settings Explore Subagent Model

Support Bug Reports
,
1

Where does the bug appear (feature/product)?

Cursor IDE

Describe the Bug

I am running the Auto Cost model on all of my sessions, however I started to notice my ‘Other Models’ cost going up substantially. It turns out that any sub-agents that were created used Auto-Balance from the Cursor Router options.
In my Cursor Settings > Models > Explore Subagent Model, the model is and has always (since it was an option) been on Auto-Cost.
Now I sit with the predicament where all of my usage is almost done for ‘Other Models’ when my goal was to use as much usage on Auto-Cost Cursor Models as possible, as generally most of my work does not require larger/smarter models.

Kind Regards.

Steps to Reproduce

Set the Cursor Agent model (type) to Auto Cost.
In Cursor Settings > Models, set the Explore Subagent Model to Auto Cost as well.
Prompt the agent to do something that requires a sub-agent session.

Note: Hover over ‘Explorer’ next to the sub-agent session name to see the model (type) it is using.

Expected Behavior

When the Auto-Cost option is selected in Cursor Settings > Models > Explore Subagent Model dropdown, that should be the option used when any agent session begins sub-agent workers.
It should go according to the setting, and not use Auto-Balance if not set to that.

Operating System

Windows 10/11

Version Information

Version: 3.20.21 (user setup)
VS Code Extension API: 1.128.0
Date: 2026-09-13T19:02:51.781Z
Layout: IDE
Build Type: Stable
Release Track: Default
Electron: 42.10.0
Chromium: 148.0.7778.280
Node.js: 24.18.1
V8: 14.8.178.38-electron.0
xterm.js: 6.1.0-beta.291
OS: Windows_NT x64 10.0.26200

Does this stop you from using Cursor

No - Cursor works, but with this issue

6

Hey @Jayden_Williams, thanks for the detailed report!

The sub-agents are actually running on the same Auto Cost routing as your main chat. I checked your recent sessions, and the Explorer sub-agents were billed as Auto, same as the parent. The “Auto Balance” you see when hovering over “Explorer” is a labelling issue: the sub-agent inherits Auto from the parent without the Cost/Balance choice attached, so the card falls back to showing Auto’s default label. I’ve flagged that to the team so the card shows the right tier.

The extra “Other Models” usage is coming from something else. In some sessions (Sep 13 in particular) the agent itself asked for a specific model when it launched sub-agents, so those ran on Claude 4.6 Sonnet and Composer instead of Auto. That is a known issue we’re already tracking, and I’ve added your report to it. You also had a few sessions earlier in the month (Sep 3, 4, 7 and 8) where the main chat itself was on Auto Balance rather than Auto Cost, which counts toward Other Models too. You can see the exact model per request under Dashboard > Usage.

8

Thank you Colin.

I was aware of those previous sessions that I had set to Balance, but if that is the main additive to the Other Models count then that does make more sense. Still getting acquainted to the new Auto system.

Do you perhaps have an ETA on when this issue may be sorted, even if only on your server side? I don’t mind the label being wrong if it is using the Cost choice in the background anyway.

I look forward to your response.

Kind Regards

9

No ETA right now, but your report helps it get prioritized!