When Copilot chooses the models, an administrator still chooses which models it may use.
GitHub’s October 28 Copilot Fusion announcement makes automatic model routing and orchestration generally available across all Copilot plans in Visual Studio Code, the Copilot app, Copilot CLI and JetBrains IDEs. Users select Standard, High or Max effort rather than a specific model.
Fusion can send a task to one model, have an efficient model draft before a quality check decides whether to escalate, or have another model critique a draft before revision. That changes the unit being selected: a request can invoke a small workflow whose cost and behavior depend on several calls.
The announcement says Fusion respects enabled enterprise and organization model policies. Administrators can also disable a model specifically for Fusion while leaving it available elsewhere, and can set Fusion’s effort level as the default through managed settings. Those controls let an organization define the permitted pool before asking the router to choose within it.
Billing follows every model used at its standard rate. GitHub is offering paid plans a temporary 30% discount on model costs through Fusion. A launch discount and an efficient workflow affect the bill in different ways. Keep the undiscounted comparison visible when deciding whether the routing system should become the default.
For a pilot, use representative coding tasks with the same acceptance criteria and review process. Record effort level, observable model calls, billed usage, elapsed time and whether the change passed review. If a complete route is unavailable to the administrator, retain that visibility limit with the result rather than assuming which models participated.
A useful trial ends with a change someone can accept and a cost someone can explain. The model picker has become an operating decision; the organization still needs a clear account of what it authorized and what useful work it received.