GitHub Copilot Fusion automates model selection to cut API costs in half

GitHub introduced Copilot Fusion on Wednesday at GitHub Universe in San Francisco, positioning the new offering as a model selection and orchestration layer that removes decision-making burden from developers during coding work.

The system merges capabilities from Project HydraFusion with Copilot Auto Model Selection to handle both model routing and orchestration automatically. GitHub reports that Copilot Fusion Max achieved performance comparable to Claude Code with Opus 5.5 Max while consuming roughly half the average API cost per task, using approximately half as many tokens, and completing tasks in about 40% less time.

How Copilot Fusion makes routing decisions

Rather than requiring developers to manually select models for reasoning, code generation, debugging, or tool use tasks, Copilot Fusion evaluates requirements and routes work accordingly. The system can employ different orchestration patterns: a single model handling the entire task, a "cascading" workflow where quality checks determine whether to accept or escalate results, or a "critique" workflow where one model reviews another's output.

This approach addresses a practical problem in development workflows. As Hari Pokhrel, principal software engineer at AT&T, stated in a GitHub-supplied statement: "Previously, developers would often choose a more powerful model for a complex task and forget to switch back for everyday activities."

What developers control

While Copilot Fusion manages model selection and orchestration, developers retain control over effort level, selecting among Fusion Standard for routine coding, Fusion High for more complex reasoning tasks, and Fusion Max for the most demanding work.

Enterprise administrators gain additional controls: they can restrict which models Copilot Fusion can access without affecting model availability elsewhere, adjust policy settings per model, and determine whether Copilot Fusion serves as the default for their user base.

Broader industry shift toward automated routing

Copilot Fusion represents part of GitHub's broader strategy to position itself as "the most comprehensive software delivery system in the world," with developers delegating model choices while maintaining responsibility for work direction and outcomes. The company unveiled the system alongside other products including the GitHub Copilot app, Copilot Projects, and Copilot Automations.

Automated model routing is gaining traction across the industry. Ramp moved model selection into Open SWE, its open-source coding agent, this month. Earlier in the summer, Replit made its "intelligent model routing" system the default across all accounts.

According to Pokhrel's statement, Copilot Fusion Standard has improved token efficiency at AT&T by reducing unnecessary spending and preserving capacity for higher-value work. GitHub reports that on the Artificial Analysis Coding Agent Index, Copilot Fusion Max scored 60, placing it within two points of Claude Code with Opus 5.5 Max.