Request is going to another AI model that we didnt selected

Bug: Model selection appears to leak between CLI tabs

Environment

• Command Code CLI: v1.26.0
• OS: Windows 10

Description

I noticed an issue where the model selected in one Command Code CLI tab appears to affect another already-running tab.

My workflow was:

1. I had a main Command Code tab using Muse Spark 1.2 Contributor for my main implementation work.
2. I opened a second CLI tab and temporarily selected GLM 5.3 to review some code.
3. I made only 2 prompts in the GLM 5.3 tab because of the higher cost, then closed that tab.
4. I continued working in the original tab, which was still configured to use Muse Spark 1.2 Contributor.
5. However, when I checked the usage logs afterward, every request made after using the second tab was attributed to GLM 5.3, even though I continued using the original Muse Spark 1.2 tab.

This resulted in approximately $8–$9 of unexpected usage today.

Expected behavior

The model selection should remain isolated to the respective CLI session/tab. The original tab should continue using Muse Spark 1.2 Contributor after the second tab is closed.

Actual behavior

Selecting GLM 5.3 in a second CLI tab appears to have changed the model used by the original tab as well, causing subsequent requests from the original tab to be processed/recorded as GLM 5.3.

Impact

This can silently switch users to a significantly more expensive model without them realizing it, resulting in unexpected usage costs.

Could you please investigate whether model selection/state is being shared between CLI sessions/tabs in v1.26.0, and why subsequent requests from the original Muse Spark 1.2 session are being routed to GLM 5.3?

I can provide the usage logs/screenshots showing the model selection and request history.

Expected Behavior

nil

Actual Behavior

nil

Steps to reproduce the issue

nil

Command Code Version

Unknown

Operating System

Windows

Terminal/IDE

Unknown

Shell

cmd.exe

Session file (optional)

No response

Fix prompt (optional)

No response

Additional context

No response