GitHub Copilot Auto now lets developers choose whether routing optimizes for cost or intelligence

On September 14, 2026, GitHub added three selectable optimization tiers to Copilot Auto model selection: Efficiency, Balance and Intelligence. Efficiency prioritizes lower cost for straightforward work, Balance weighs cost, quality and latency together, and Intelligence prioritizes quality for more complex tasks. All three tiers draw from the same set of models and Auto still evaluates every prompt individually. GitHub says the feature is rolling out across Visual Studio Code, Copilot CLI and the Copilot app. Usage is billed at the rate of the model Auto actually selects, and paid subscribers keep the existing 10% Auto discount.

Automatic model routing is becoming part of coding-agent economics rather than just an invisible vendor implementation detail. Until now, Copilot Auto largely decided the cost-versus-capability tradeoff on the user's behalf. The new tiers expose that tradeoff as a user-controlled setting without requiring developers to pick a specific model for every prompt. That can make mixed workloads easier to manage: routine edits can stay cost-sensitive while harder debugging or architectural work can bias toward stronger models. The important caveat is that selecting Intelligence does not force a premium model, and selecting Efficiency does not set a hard spending cap; billing still depends on the model chosen for each prompt.

GitHub says Efficiency, Balance and Intelligence all use the same pool of models available to the user's plan and policies. The difference is how Auto weighs cost, quality and response time. A simple request can still be routed to a smaller model even under Intelligence if GitHub judges that model sufficient.

The tier itself does not have a fixed price. Copilot charges according to the model Auto selects for the prompt. Paid subscribers continue to receive the existing 10% discount for usage routed through Auto, so developers still need to watch underlying model consumption rather than treating Efficiency as a budget ceiling.

GitHub had already made Auto task-aware in VS Code and Copilot CLI, evaluating factors such as reasoning complexity, code generation, bug diagnosis and tool orchestration. The September change adds an explicit user preference on top of that routing system rather than replacing per-prompt evaluation.

GitHub describes the three tiers as the first step toward customizable model selection with greater visibility into tradeoffs. That suggests model choice in Copilot may increasingly resemble a policy layer where users and organizations specify objectives while the platform handles individual routing decisions.