GitHub announces official release of Copilot Fusion
November 6, 2026 · 2 min read
Also published in Nederlands
GitHub has announced the official release of its new feature, GitHub Copilot Fusion, during the annual event GitHub Universe 2026 held on October 28 and 29 in San Francisco. This feature is designed to optimize the selection of AI models based on the specific characteristics of tasks, allowing for the execution of a single or multiple AI models in a multi-model setup.
Copilot Fusion evaluates the necessary capabilities for different tasks such as inference, code generation, debugging, and tool utilization, determining how to best combine AI models for these purposes. This aims to reduce the burden on developers to manage the selection of AI models individually while facilitating development that balances performance and cost.
The feature offers three patterns for task execution, depending on the task at hand. One pattern enables an AI model selected for efficiency to generate solution proposals, which a quality assurance AI model then verifies, deciding whether to escalate to a higher capacity AI model. Another pattern involves one AI model drafting a solution, which is independently reviewed by an AI model from a different family. When feedback is returned from the review, the original AI model can make a single revision based on that feedback.
To help balance quality and cost, different load levels are provided for selecting among the patterns. A medium token-saving approach offers balanced orchestration suitable for large coding tasks with clearly defined scopes, such as complex bug fixes and changes spanning multiple files. Another approach prioritizes cutting-edge performance while considering token efficiency, aimed at tasks requiring deep inference and thorough analysis. A scenario requiring a fast model can be chosen, although users cannot specify it.
Additionally, as part of the public preview, users can execute tasks using locally installed MAI Code 1.1 Flash or Qwen3.8 Flash Next models within the Fusion Standard framework.