effortLevel config is not respected when switching models via /model command

[juchao-neoncrm]

Describe the feature or problem you'd like to solve

No response

Proposed solution

The effortLevel setting in ~/.copilot/config.json is correctly applied when a new session starts. However, when switching models mid-session using the /model command, the effort level resets to "medium" by
default instead of inheriting the value from the config.

Expected behavior: When switching models via /model, the effort level should default to the value set in config.json (effortLevel), so users don't have to manually re-select it every time.

Suggested improvement: Either:

1. Pre-select the configured effortLevel in the /model effort picker UI, or
2. Support per-model default effort levels in config.json (e.g., "models": { "claude-sonnet-4.6": { "effortLevel": "high" } })

Steps to reproduce:

1. Set "effortLevel": "high" in ~/.copilot/config.json
2. Start a new session — effort level is correctly "high" ✅
3. Use /model to switch to another model — effort level resets to "medium" ❌
4. Use /model to switch back — still "medium" ❌

Example prompts or workflows

No response

Additional context

No response

[NotAProgrammer97]

I can reproduce this on Windows with Copilot CLI 1.0.34.

Environment

• Windows
• Copilot CLI 1.0.34
• config contains "effortLevel": "high"

What I verified locally

1. A fresh plain Copilot CLI session still inherits high correctly.
   The process log resolved the model as:
   capi:gpt-5.4:defaultReasoningEffort=high

2. The problem appears after model selection, not at startup.
   When the model is changed through /model, the effort falls back to medium instead of preserving the configured high value.

3. I also tested explicit GPT-5.4 xhigh launch commands on 1.0.34, and they still resolved to medium:
   copilot --model gpt-5.4 --reasoning-effort xhigh -p "Reply with OK."
   copilot -p "Reply with OK." --model gpt-5.4 --reasoning-effort xhigh

   In both cases the process log showed:
   capi:gpt-5.4:defaultReasoningEffort=medium

So on this machine there seem to be two related problems:

• /model does not preserve the configured effortLevel
• explicit xhigh selection for GPT-5.4 may also be ignored and collapse to medium

This feels related to #2725 and #2739, but the key point here is that startup is not the main failure. Startup is high. The regression happens on model selection, and explicit xhigh also does not stick in my tests.