GitHub Copilot Adds Auto Model Selection Tiers
GitHub CopilotGitHub Copilot's auto model selection now offers three configurable tiers, Efficiency, Balance, and Intelligence, letting developers decide how auto should weigh cost, quality, and response time for each prompt. All three tiers draw from the same pool of available models, and Copilot still evaluates every prompt individually, so a simple task can still route to a small, fast model even under the Intelligence tier. The feature is rolling out in Visual Studio Code, Copilot CLI, and the GitHub Copilot app, with usage billed according to whichever model auto actually selects and paid subscribers still receiving a 10% discount on auto-billed usage.
Key Takeaways
- GitHub Copilot's auto model selection now exposes three named tiers, Efficiency, Balance, and Intelligence, instead of a single fixed cost/quality weighting.
- All three tiers draw from the same pool of models; the tier only changes how auto weighs cost, quality, and latency per prompt, it does not restrict which models are available.
- Even in the Intelligence tier, simple tasks like adding a docstring can still route to a small, cheap model, since auto still evaluates every prompt individually.
- The Efficiency tier is aimed at fast, low-stakes tasks, while Intelligence is aimed at complex work where output quality matters most.
- Billing is tied to whichever model auto actually selects, not to the tier chosen, and paid subscribers keep their existing 10% discount on auto-billed usage.
- The rollout spans VS Code, Copilot CLI, and the GitHub Copilot app, and GitHub describes it as the first step toward deeper customization of auto model selection.
A New Way to Tune Auto Model Selection
GitHub Copilot's auto model selection feature, which automatically picks the best available model for a given prompt, now gives developers direct control over how that decision gets made. Instead of a single, fixed weighting between speed, cost, and output quality, Copilot now exposes three named tiers: Efficiency, Balance, and Intelligence. Each tier changes how aggressively auto optimizes for cost versus quality when it evaluates an incoming prompt.
The Three Tiers
The Efficiency tier prioritizes keeping costs low and is best suited to fast, straightforward tasks where a smaller model gets the job done just as well. The Balance tier weighs cost, quality, and latency together, and is designed as the default-feeling choice for everyday development work. The Intelligence tier prioritizes output quality above cost or speed and is aimed at complex, higher-stakes tasks where getting the best possible result matters more than saving on usage.
Importantly, GitHub notes that all three tiers pull from the same underlying set of models. Choosing a tier does not lock a developer into a fixed model; instead, it changes the weighting auto applies when it independently evaluates each prompt. As a result, even under the Intelligence tier, a trivial task like adding a docstring to an existing function may still be routed to a small, efficient model, since auto still makes a fresh, per-prompt judgment rather than blindly escalating to the most expensive option available.
Availability and Billing
The tiered auto model selection is currently rolling out across Visual Studio Code, Copilot CLI, and the standalone GitHub Copilot app. Billing is unaffected by which tier a developer selects: usage is charged based on whichever model auto actually chooses for a given prompt, not based on the tier itself. Paid subscribers continue to receive the existing 10% discount on usage billed through auto, regardless of tier.
GitHub frames this release as the first step toward giving developers more visibility into, and control over, the tradeoffs auto model selection makes on their behalf, with further customization expected to follow.