Auto chooses and charges Sonnet 5

Where does the bug appear (feature/product)?

Cursor IDE

Describe the Bug

I issued a prompt in Auto mode and the agent chose my last chosen Sonnet 5 model.

Steps to Reproduce

Enter prompt to plan a feature set on Auto mode.

Expected Behavior

Auto mode should have been chosen, it chose Sonnet 5 Medium No Thinking which is a model I used a few turns ago in a different chat prompt.

Screenshots / Screen Recordings

Screenshot From 2026-07-16 00-21-40.png

Operating System

Linux

Version Information

Version 3.11.25

For AI issues: which model did you use?

Auto

Does this stop you from using Cursor

No - Cursor works, but with this issue

What you’re seeing isn’t the main Auto request running Sonnet 5, it’s the Explore step. When you plan on Auto, the agent can spin up a short Explore subagent to research your codebase first, and subagents choose their own model instead of following the Auto picker. That research subagent ran Sonnet 5, which is why the usage showed up even though the picker still reads Auto.

This is a rough edge we’ve seen reported, and we’re improving how subagents pick models so they don’t quietly land on a premium one. For now you can control it directly: in Cursor Settings → Models, set the Explore subagent model to a lighter option (a fast Composer model is a good default) so the research step stops using Sonnet 5.

Your main Auto selection is working as expected here; it’s just that the subagent step carries its own model. Let me know if changing that setting doesn’t stick and I’ll dig in further.