What happened

GitHub posted a changelog entry on September 14. A changelog is the running list GitHub keeps of product changes. This one announces three tiers for Copilot’s auto model selection.

Auto is the mode where Copilot picks which underlying AI model handles a prompt for you. You don’t choose a model by name every time. Now you also choose how auto should weigh cost, quality, and speed while it picks.

Efficiency favors low cost and suits fast, simple tasks. Balance weighs cost, quality, and speed together for everyday work. Intelligence favors quality for complex tasks.

All three tiers draw from the same pool of models. Copilot still evaluates each prompt on its own and picks whatever model fits, regardless of tier. GitHub’s own example: adding a one-line docstring, a short comment describing what a function does, is a simple task. Even on the Intelligence tier, auto can still route it to a smaller, cheaper model. The task doesn’t need more than that.

The rollout is live now in Visual Studio Code, Copilot CLI, and the GitHub Copilot app. Copilot CLI is Copilot’s command-line tool, for people who trigger it by typing commands rather than clicking through an editor. GitHub’s documentation describes this as a gradual rollout still reaching users, not a single finished release. Neither the changelog nor the docs name the exact menu or settings screen where you switch tiers in each surface. If you don’t see the option yet, that’s consistent with a rollout still in progress.

Why it matters

The part worth slowing down on is billing. GitHub’s changelog states usage is charged based on whichever model auto actually selects, regardless of the tier you picked. Paid subscribers keep a 10% discount on usage billed through auto, on top of that.

That means picking Efficiency is a preference, not a spending cap. It raises the odds that auto leans toward cheaper models. But a task auto judges as demanding can still route to an expensive model, even on the Efficiency tier. You’ll be billed for that model’s usage either way. If you were hoping a tier setting works like a monthly budget limit, it doesn’t. It’s a bias dial on model choice, not a ceiling on spend.

The flip side matters just as much if you’ve been avoiding auto mode because you didn’t trust it to pick well for demanding work. Intelligence tells auto to bias toward quality specifically for the tasks that need it. You still don’t have to hand-pick a model for every prompt.

Who should care

Anyone using Copilot’s auto mode in VS Code, the Copilot CLI, or the Copilot app. That matters most if you or your team track Copilot usage against a monthly allowance or budget. It also matters if you’ve stuck to manually picking a specific model every time because auto felt unpredictable. These tiers are the first real lever GitHub has given you over how auto behaves, short of turning it off.

What builders should do next

Don’t assume Efficiency caps your costs. Test it. Pick five prompts you send regularly. Mix in small edits and something bigger, like refactoring a function or writing a test.

Run all five once on Efficiency and once on Intelligence. Note which model actually handled each one (Copilot surfaces this in its response). Compare how the output quality differed for the harder prompts.

If Efficiency routes your demanding prompts to the same model Intelligence does, the tier isn’t saving you anything on those tasks. Switch to whichever tier gives you the response quality you want by default instead. If cost is what you actually care about, watch your Copilot usage total over a week on each tier. Don’t just trust the tier name alone.

GitHub hasn’t published exactly where the tier switch lives in each surface yet. Check the model picker or settings menu in VS Code, the CLI, or the Copilot app directly. Or follow GitHub’s auto model selection docs for current instructions once your rollout lands.

End of article