Automatic switching between model for planning and execution

[hakontel]

Describe the feature or problem you'd like to solve

I want Copilot to be able to use one configurable model for the planning step of a task, then automatically switch to another model for the execution of that plan

Proposed solution

It will be more efficient for everybody if this option is available. Currently achieving this is a manual changeover and something I suspect many (like me) are not willing to go back and forth in everyday use. The benefit is using less resources on achieving the same thing, for example if you want Opus 4.6 to draft the plan on what to do, but the execution can be relegated to Sonnet 4.6 then we consume less energy to achieve the same thing as just using Opus for both parts of the task.
The benefit for Copilot CLI is an attractive feature its users will appreciate and allow competition with for example Claude Code

Example prompts or workflows

considering the change would be a configuration possibility the prompts and workflows should remain the same, just utilizing different models under the hood

Additional context

Let the user configure Copilot CLI to enter a multi-agent mode where the user can configure the following:

• select from available models to use during planning phase
• select from available models to use during execution phase

[lucazav]

That’s exactly how Cline works (Plan + Act) , and it’s really convenient because you can assign a different model to each of the two modes. It would be great to have this option in Copilot CLI as well

[wiltaylor-daf]

This is going to become more of a necessity with the billing changes happening 1 June.

[magnusakselvoll]

This is one of the big shortcomings to me coming from Claude Code. My experience is that the best place to budget tokens is on the plan; if the plan is good the implementation is easier. So I typically use the best available model to create the plan, revise it properly, and then implement using one of the more budget oriented models.

In Claude this is easily done with /model opusplan, but I presume something a bit more explicit like the VSCode settings chat.planAgent.defaultModel + github.copilot.chat.implementAgent.model would be more the way to go.

Really hoping this issue will get some traction, as knowing how to budget your AI tokens is one of the most important skills moving forward.

[newtob]

This would be a highly valuable feature.