GitHub made Copilot Fusion generally available on October 28 at Universe. You select Copilot Fusion in the model picker, choose an effort level (Standard, High or Max), and Copilot decides per prompt which models run. It can send a task to one model, let a cheap model draft and escalate to a stronger one if a quality check fails (GitHub calls this Cascade), or have one model draft while another reviews it (Critique).
The numbers come from Artificial Analysis. On its Coding Agent Index, Fusion Max scored 60, two points behind Claude Code with Opus 5.5 Max. VentureBeat reports the average cost per task as $5.59 against $11.42, and GitHub says Fusion Max also took about 40% less time per task. A higher "up to 77%" savings figure that Microsoft has cited comes from a separate company-run evaluation.
Billing: you pay each model Fusion uses at its standard rate, and paid Copilot plans get 30% off model costs when using Fusion for a limited time. Admins can restrict which models Fusion may pick and make it the default. It works in VS Code, the GitHub Copilot app, Copilot CLI and JetBrains IDEs. GitHub also announced Copilot projects (agent work across several repos in parallel sessions) and session continuity, so a cloud agent session started on your phone can be picked up in VS Code, the CLI or github.com.

