Skip to content
NewsProduct launch

GitHub Copilot Adds Efficiency, Balance, and Intelligence Tiers to Auto Model Selection

· by Pondero Newsdesk

The short version

GitHub's three-tier system for Copilot auto model selection lets developers set a cost-quality preference that applies automatically on every prompt, with no manual model switching.

GitHub Copilot Adds Efficiency, Balance, and Intelligence Tiers to Auto Model Selection

GitHub shipped three configurable tiers for Copilot auto model selection on September 14. Developers can now tell Copilot how to weigh cost, quality, and response time on every prompt, replacing a single opaque default with an explicit preference they set once and forget.

What shipped

Per the GitHub Changelog, the three tiers are efficiency, balance, and intelligence. Efficiency prioritizes keeping costs low and suits fast, straightforward tasks. Balance weighs cost, quality, and latency together and fits everyday work. Intelligence prioritizes quality for complex tasks.

All three tiers draw from the same roster of models. Auto evaluates each prompt individually and selects the best-suited model for the chosen tier. A simple task such as adding a docstring to an existing function may use a small, lightweight model even in intelligence mode. Usage is charged based on whichever model auto selects, regardless of the tier setting. Paid subscribers retain the existing 10 percent discount on usage billed through auto.

The rollout covers Visual Studio Code, Copilot CLI, and the GitHub Copilot app. GitHub described the release as "the first step" in a broader roadmap toward model-selection transparency, without disclosing a specific timeline for what follows.

Why it matters

The routing rule is simple: efficiency for high-volume boilerplate (docstrings, completions, renaming), intelligence for code review, architectural reasoning, or complex debugging. Before this release, developers had no way to express that preference. They got whatever auto decided, with no feedback on what that was.

The per-prompt auto evaluation means the tier is a preference rather than a ceiling. The efficiency setting will still pick a stronger model when the prompt demands it. A team generating thousands of completions per day across a mix of trivial and complex tasks can route the bulk through efficiency and reserve the intelligence tier for work that warrants it. The 10 percent paid-subscriber discount on auto usage compounds across both tiers.

GitHub's "first step" framing points to follow-on work: specifically, more visibility into the cost tradeoffs a developer is making prompt by prompt. That transparency would let developers and managers connect tier choices to actual token costs, a capability the current release does not yet deliver.

What to watch next

Two items on the public roadmap will shape how useful this becomes at the org level. Per-tier token usage breakdowns in the Copilot usage metrics dashboard would let engineering managers measure the actual cost impact of tier adoption across their team. Organization-level tier defaults for enterprise admins would convert this from a per-developer preference into a cost-management policy. Neither is available yet.

The Copilot auto model selection documentation covers how auto evaluates prompts and lists the models available through auto.

Sources