GitHub Copilot Auto adds three ways to steer cost and quality

GitHub Copilot Auto now lets developers choose whether its model routing should favor lower cost, an everyday balance, or higher response quality. Announced on September 14, the new Efficiency, Balance, and Intelligence preferences add practical control without requiring users to select a model for every prompt.

Three preferences, not three fixed models

Efficiency prioritizes lower model costs for fast, straightforward work. Balance weighs cost, quality, and response latency together for everyday tasks, while Intelligence prioritizes response quality for more complex requests.

Choosing a preference does not pin Copilot to a particular model. Every tier uses the same eligible model set available to the user, and Auto still evaluates each prompt before routing it. Even under Intelligence, a simple request such as adding a docstring may go to a smaller model. Task complexity, real-time model health, and availability also inform the choice.

The selected model sets the cost

Usage is billed according to the model Auto actually selects, regardless of the routing preference. Paid subscribers continue to receive the existing 10% discount on model costs when using Auto. Eligible models depend on the user’s Copilot plan and administrator policies and can change over time.

The tier picker is currently rolling out in Visual Studio Code, Copilot CLI, and the GitHub Copilot app. Task-aware Auto routing is available more broadly in Copilot Chat on GitHub.com and supported products, but the three preference controls are limited to those three surfaces for now.