September 15, 2026: GitHub announced new GitHub Copilot auto model selection tiers on September 14. Users can choose how automatic routing weighs cost, response speed and quality. The feature is rolling out in VS Code, Copilot CLI and the GitHub Copilot app, according to the official announcement.
What the three tiers change
Efficiency favors lower cost for straightforward work. Balance considers cost, quality and latency together. Intelligence emphasizes quality for complex tasks.
Each tier draws from the same available model pool; Auto evaluates the prompt to choose a model. Selecting Intelligence therefore does not guarantee a larger model for every request. Charges follow the model selected, and paid subscribers retain the 10% discount on usage billed through Auto.
Availability and traceability
GitHub’s Auto documentation says model access remains subject to subscription and administrator policies. Teams can inspect the model used for a response: Copilot CLI displays it in the terminal, while Copilot Chat exposes it on hover. The broader availability of Auto should not be confused with completion of this new tier rollout.
Why this matters for QA engineers
QATechTools recommends evaluating these preferences against the work your team actually delegates. A test-comment rewrite and a difficult regression investigation have different success criteria. The new choices create a useful comparison point, but the announcement does not establish better defect detection or more reliable tests.
For a small pilot, repeat a fixed set of tasks across the available tiers: explain a failing assertion, propose boundary cases and repair a deliberately broken test. Keep repository state and prompts consistent. Record the selected model, elapsed time, usage cost and whether the result passes independent checks. These are suggested QA checks, not GitHub benchmark results.
Choose a preference from measured outcomes, and retain normal review and regression gates for generated changes. Recheck results when the available model pool changes; a tier name alone is insufficient evidence of reproducible behavior.