GitHub shipped an update on September 14 that gives developers more control over how Copilot picks a model automatically. Auto model selection, which already decided on its own which model to use for each prompt, now accepts three preference tiers: efficiency, balance, and intelligence, according to the official changelog.
Three tiers, the same set of models
The difference between tiers is not which models are available — all three draw from the same pool — but how Copilot weighs cost, quality, and latency when picking one for each individual prompt. Efficiency prioritizes keeping costs low and suits quick, straightforward tasks. Balance weighs all three factors together for everyday work. Intelligence prioritizes quality, built for complex tasks.
One detail avoids the obvious trap of assuming "intelligence" always triggers the most expensive model: GitHub's own example is a simple task, like adding a docstring to an existing function, that can still run on a small, efficient model even when the tier is set to intelligence. The choice happens per prompt, not per session.
Rollout and billing
The feature is rolling out in VS Code, Copilot CLI, and the GitHub Copilot app. Billing follows whatever model Auto actually selects, regardless of the chosen tier: setting "efficiency" does not change the charge associated with the model selected; it only influences which model is picked more often. Paid subscribers keep the existing 10% discount on usage billed through Auto.
Why it matters to whoever pays the AI bill
Engineering teams managing Copilot usage budgets have often treated automatic model selection as a black box, with limited control over the tradeoff among cost, quality, and latency. Exposing the efficiency/balance/intelligence dial is a direct answer to that recurring complaint, though the final per-prompt call still sits with GitHub's algorithm, not the developer. It is worth watching how the "efficiency" tier actually behaves over a few weeks before assuming it solves usage-cost predictability on its own: a preference dial is not the same as picking the model by hand, and the difference between the two only shows up on the bill at the end of the month.