Copilot auto model tiers: you pick the dial, not the model

GitHub shipped a cost dial for Copilot on September 14. Auto model selection now takes a preference — Efficiency, Balance, or Intelligence — and routes each prompt to a model accordingly. It's a genuinely useful control. It is also a reminder that in a routed product, you no longer choose the model; you choose a posture, and you get billed for whatever the router decides.

What actually happened

Per the GitHub changelog, the three tiers are straightforward: Efficiency prioritizes keeping costs low and suits fast, straightforward tasks. Balance weighs cost, quality, and latency together for everyday work. Intelligence prioritizes quality and is built for complex tasks.

Two details matter more than the names. First, auto evaluates each prompt individually even within a tier — GitHub's example is that a simple documentation addition may route to a smaller model regardless of which tier you selected. The tier is a bias, not a floor. Second, on billing: "Usage is charged based on the model auto selects, regardless of tier." Paid subscribers continue to get a 10% discount on usage billed through auto.

The feature is rolling out in Visual Studio Code, Copilot CLI, and the GitHub Copilot app.

Why model routing matters for your business

A dial is not a budget. "Efficiency" biases toward cheaper models; it does not cap your spend. If a task needs six turns on a large model, Efficiency mode may still spend them and you will still be billed. Treat the tier as a hint to the router, and keep whatever actual spending limit your plan offers as the thing that stops the bleeding. Those are different controls and only one of them is enforcement.

Measure cost per finished task, not cost per request. This is the mistake we watch teams make with every routed AI product. Efficiency mode looks cheaper per call and can be more expensive per outcome, because a weaker model takes more turns, produces more code to review, and occasionally sends a developer down a dead end. The only number that means anything is what it cost to get a working, reviewed, merged change. Run a week on Balance and a week on Efficiency against similar work and compare the merged output, not the invoice line.

Pick tiers by task, not by team. The honest split for most small shops: Efficiency for boilerplate, renames, test scaffolding, and docs. Intelligence for anything touching money, auth, data migrations, or a system you don't fully understand yet. Balance for everything else. That's a decision you can write on one line and hand to a new hire.

Routing means your toolchain can change without you. The model behind a given prompt today is not necessarily the model behind it next month — vendors add, retire, and re-rank models continuously, and a router absorbs that silently. That's convenience, and it's also drift. Keep the parts you control portable: your prompts, your review process, your evals. We've written before about pinning your model layer — routing tiers make that discipline more useful, not less.

Source: GitHub Changelog, September 14, 2026.