What are the three Copilot auto model selection tiers?

Efficiency (low cost, fast tasks), Balance (mix of cost, quality, latency), and Intelligence (quality first, complex tasks). All three tiers use the same available model set; the tier only biases how auto picks a model for each prompt.

Does the tier change Copilot pricing?

No. Usage is charged based on the model auto selects, regardless of tier. The only published discount is the existing 10% for paid subscribers on usage billed through auto. No per-tier pricing or savings figures were released.

Where is Copilot auto model selection available?

The tiered selection is rolling out in Visual Studio Code, Copilot CLI, and the GitHub Copilot app. JetBrains and the IDE plugins are not included in this release.

GitHub Copilot Auto Model Selection Gets Efficiency, Balance, Intelligence Tiers

GitHub shipped three selectable tiers — efficiency, balance, intelligence — for Copilot's auto model selection, rolling out now in VS Code, Copilot CLI, and the Copilot app.

What changed?

Auto model selection now exposes a dial: efficiency biases toward cheap and fast, intelligence biases toward quality, balance mixes all three. Two details matter more than the tier names. First, all three tiers draw from the same model set — the tier only re-weights per-prompt selection, and GitHub says even the intelligence tier will pick a small model for a docstring task. Second, billing is unchanged: you pay whatever model auto selects, per its standard rate. The only economics GitHub published is that paid subscribers keep a 10% discount on auto-billed usage.

Why a builder cares?

Read the fine print: the tier names sound like cost controls, but there is no per-tier price, no per-tier cap, and no published savings data. You cannot compute what "efficiency" saves versus "intelligence" from anything GitHub shipped today — you can only find out on your bill. Compare that to OpenRouter's handling of the same routing problem this week: every alias model on OpenRouter publishes its exact rate in the public models API (for example, DeepSeek Flash at $0.15/$0.60 per million tokens), so a routing redirect is auditable before it hits your invoice. GitHub's tiers are a bias knob with a 10% discount bolted on; budget-constrained teams should treat tier selection as unmeasurable until GitHub exposes spend-per-tier in Copilot usage metrics — which, notably, it started surfacing for VS Code agents on September 11. The gap between "choose your tradeoff" and "see your tradeoff" is where costs hide.

Also absent today: JetBrains and the IDE plugins. The tier control exists only in VS Code, CLI, and the Copilot app.

What changed?

Why a builder cares?

Related