Cloud Agents subagents use different models than selected

[hadrian]

Where does the bug appear (feature/product)?

Cloud Agent (GitHub, Slack, Web, Linear)

Describe the Bug

Cloud Agents subagents are using different models (like Claude Sonnet, and Google Gemini) than the one selected for the task (like Grok 4.6). I discovered this when investigating why my “Other Models” usage was depleting while I was always selecting Grok 4.6.

THIS IS WRONG! Doesn’t matter which model is best for the task, it is wrong, if I only enabled Grok 4.6, then only Grok 4.6 should be used for everything. I was not informed that other models would be used, and I only have Cursor Grok 4.6 enabled in the Models settings (I know this is for the model picker, but it should be for everything).

I intentionally paid for Cursor subscription to use Grok and Cursor’s Cloud Agents. If this isn’t fixed I’ll migrate to a different solution because using other models with Cursor, instead of Grok, is not worth it given there’s no additional usage beyond what’s paid.

Please fix this, otherwise Cursor Cloud Agents become uneconomical for complex tasks.

Steps to Reproduce

Start a Cloud Agents tasks for browser extension development. Instead of using the selected model for everything, other models will be used for subagents

Expected Behavior

Only models enabled in the Models settings are selected for Cloud Agents subagents

Screenshots / Screen Recordings

Operating System

MacOS

Version Information

Cursor Cloud Agents (Aug 30 2026)

Does this stop you from using Cursor

Sometimes - I can sometimes use Cursor

[deanrie]

Hey, thanks for the detailed report with screenshots. I can see the usage breakdown showing Sonnet and Gemini next to Grok.

Here’s what’s happening. The main Cloud Agent really did run on Cursor Grok 4.6, as you selected. The Claude Sonnet and Gemini lines in usage are helper subagents that the agent spawns for support tasks like browser checks, screenshots, and video analysis. Right now, some of these helpers pick their own model instead of inheriting the one you chose.

The list in Settings → Models only filters the model picker in the IDE. It doesn’t yet limit which models can be used for cloud subagent tasks, so disabling other models didn’t help.

This is known behavior we’re tracking, and it’s already been shared with the team. I can’t give a timeline yet, but there’s a related thread you can follow here: Cloud Agent logs claude-4.5-sonnet when picker is Grok 4.6 and Claude is not enabled

I get that unexpected spend under Other Models is frustrating. If you want to dig into a specific charge or usage on your account, it’s best to email [email protected] so they can check account-specific details. Let me know if you’ve got questions.