At GitHub Universe 2026, GitHub officially launched Copilot Fusion, an intelligent orchestration layer that automatically routes and combines multiple AI models from OpenAI, Anthropic, Google, and others to complete coding tasks, without requiring developers to manually choose a model. The feature combines existing automatic model selection technology with the multi-model orchestration capabilities of Project HydraFusion, introduced in a September research preview.
Developers only need to select Copilot Fusion in the model selector. The system then automatically decides, based on task complexity, whether to use a single model, Cascade, or Critique execution mode, and matches it to one of three effort levels: Standard (focused on efficiency and cost), High (enhanced reasoning), or Max (pursuing the strongest capability), avoiding unnecessary expense for simple tasks.
On the Artificial Analysis Coding Agent Index benchmark, Copilot Fusion Max scored 60 points, just 2 points lower than Claude Code running Opus5.5Max, but with average API cost about half, token consumption cut in half, and task completion about 40% faster. The feature is now generally available in all GitHub Copilot plans, across Visual Studio Code, CLI, the app, and JetBrains IDEs, with no extra orchestration fee and billing only for actual model calls, plus a 30% discount on model fees through November 30, 2026. Enterprise administrators can control which models are available and set it as the default experience. In addition, the system supports experimental local model routing, such as calling Microsoft MAI Code1.1Flash on Windows devices.
Microsoft also launched Copilot harness, developed by the GitHub team. It already powers multiple products, including Copilot Studio and VS Code, and is available to developers through Microsoft Foundry. Multi-model intelligent orchestration is becoming an important direction in the evolution of AI coding tools, promising a better balance between performance and cost.