When i use qwen3.6-plus in copilot CLI, copilot CLI use much too tockens than in claude code.

### Describe the bug

I use qwen3.6-plus model. Copilot CLI take 40% of my tockens in 3 hours, but claude use 3% only,when do same things. But when I use deepseek in copilot CLI ,it save tockens very much such as I use qwen3.6-plus model in claude code. So, what't the matter with copilot CLI and qwen3.6-plus. I will not use copilot CLI, if it takes so much tokens.

### Affected version

_No response_

### Steps to reproduce the behavior

_No response_

### Expected behavior

_No response_

### Additional context

_No response_

Start by reproducing the reported token usage in Copilot CLI with qwen3.6-plus, then compare the same work with Claude Code and DeepSeek as described. Record the CLI version, prompts, and token counts; the issue is complete when the excess usage is explained and a focused fix or confirmed upstream cause is documented.

由索引模型根據 Issue 內容生成。