GitHub ships Copilot Fusion for automatic model routing

Builders integrating Copilot into internal workflows can enable Fusion Max on a pilot repository and compare token consumption and task completion time against their current manual model choices. An illustrative scenario is routing a multi-step refactoring task through Fusion Max while logging the sequence of models selected and the final review time. The evidence shows a reported 40 percent time reduction and half the tokens versus Claude Code Opus 5.5 Max, yet does not disclose the exact models chosen by the orchestrator or whether the savings hold when the task set includes only simple edits.

Operators evaluating Copilot licensing should confirm with their GitHub account team which models remain selectable after any enterprise policy restrictions are applied. An illustrative scenario is an administrator disabling the most expensive frontier model for Fusion while leaving it available for direct Copilot use, then measuring aggregate spend over a two-week period on the same ticket volume. The supplied evidence establishes the existence of per-model policy controls but does not report measured cost impact or whether Fusion usage counts against existing Copilot seat limits.