GitHub launches Copilot Fusion to route coding tasks across AI models at lower cost

At its GitHub Universe conference on October 28, GitHub unveiled Copilot Fusion, a runtime orchestration engine that routes software development requests across multiple AI models and coordinates several models to complete a task.

Rather than a single model, Fusion decides whether to send a request to one model or coordinate multiple models in patterns described as single, cascade and critique. It offers three reasoning efforts — standard, high and max — and is designed to honor enterprise and organization model policies.

Independent testing by Artificial Analysis, published October 28, found that Fusion Max averaged $5.59 per task on the firm's Coding Agent Index, roughly 51% less than the $11.42 average cost of Claude Code running Anthropic's Claude Opus 5.5 at its Max setting. Fusion scored 60 overall, about two points behind the Opus 5.5 configuration, and GitHub reports it completed benchmark tasks in about 40% less time. The evaluation also measured about 7.2 million tokens per task for Fusion, roughly half the token usage of the Opus 5.5 Claude Code configuration.

Performance varied across benchmark components: Fusion led on Terminal-Bench 4.0 with a score of 71, but scored 51 on FrontierCode, behind Claude Code with Opus 5.5 at 56 and Devin Fusion at 54. Microsoft CEO Satya Nadella has cited potential savings of up to 77%, a figure based on GitHub's own DeepSWE benchmark evaluation rather than the Artificial Analysis average.

Copilot Fusion is generally available across GitHub's Copilot plans in VS Code, the Copilot app, Copilot CLI and JetBrains IDEs, with users billed at each underlying model's standard rate. GitHub is also offering paid Copilot subscribers a temporary 30% discount on model costs when using Fusion.