GitHub Copilot Fusion routes coding tasks across multiple AI models

GitHub has made Copilot Fusion generally available, turning model choice into an automated routing decision. Developers select an effort level instead of a specific model, and Fusion chooses among models allowed by their organization. It can use one model, let a cheaper model draft before escalating, or have separate models critique and revise a response.

Fusion is available on all Copilot plans in Visual Studio Code, the Copilot app, Copilot CLI and JetBrains IDEs. Standard targets everyday coding, while High and Max spend more reasoning and model calls on difficult work. Customers are billed for the underlying models used; paid plans get a temporary 30% discount on Fusion model costs.

GitHub says Fusion Max scored 60 in the Artificial Analysis Coding Agent Index, two points behind Claude Code with Opus 5.5 Max, while using about half the tokens and costing about half as much per task. Artificial Analysis reported $5.59 per task for Fusion versus $11.42 for Claude Code in that test. GitHub commissioned and funded the evaluation, so the comparison should be treated as vendor-sponsored evidence rather than a neutral verdict.

The broader shift is important: coding assistants are starting to sell orchestration rather than a single model. For teams, the product decision becomes a policy over quality, latency and budget, while the router decides which model handles each step.