Set a Copilot auto-model tier, then check which model answered

What changed

GitHub Copilot auto model selection now offers three tiers: efficiency, balance and intelligence. You choose the tier, and auto weighs cost, quality and response time for each prompt accordingly.

All three tiers use the same set of available models, and auto still evaluates each prompt. GitHub's own example: adding a docstring to an existing function may use a small, efficient model even when auto is optimising for intelligence.

What each tier prioritises

Efficiency prioritises cost and suits fast, straightforward tasks; GitHub describes it as routing prompts to the most cost-efficient and appropriately capable model.

Balance weighs cost, quality and latency together and is described as a good fit for everyday work. Intelligence prioritises quality and is built for complex tasks: a simple prompt can still go to a smaller model while a complex prompt goes to the most capable model.

The same models remain available in every tier. What changes is how preferred models are selected for each task.

Where it applies and what you pay

The changelog says the feature is rolling out in Visual Studio Code, Copilot CLI and the GitHub Copilot app. The documentation is more specific: tiers are available only on those three, while auto model selection itself is generally available in Copilot Chat on the GitHub website and supported IDEs, Copilot CLI, the Copilot app and the cloud agent.

Usage is charged based on the model auto selects, regardless of the tier you pick. Paid plans keep a 10 per cent discount on model costs while using auto model selection.

The documentation also notes that auto may serve evaluation models to users on individual plans, and that those users can disable them.

Check which model actually answered

In Copilot Chat, hover over a response to see the model used. Copilot CLI prints the model for each response in the terminal, and the cloud agent shows it at the end of the response.

A workable loop: pick a tier, run a normal week of prompts, then look at which models appeared. If complex prompts keep landing on models that miss, move up a tier; if simple prompts are landing on expensive models, move down.