At the Universe 2026 conference, GitHub officially launched Copilot Fusion—a smart orchestration layer that automatically routes and combines AI models from OpenAI, Anthropic, Google, and other providers to handle coding tasks, without requiring manual model selection by developers. This feature merges established automatic model-selection techniques with the multi-model orchestration capabilities of Project HydraFusion, which was previewed in September.
Developers simply need to enable Copilot Fusion in the model selector, and the system will automatically determine the appropriate execution mode—single model, cascade, or critique—based on task complexity. It also aligns with three effort levels: Standard (focused on efficiency and cost), High (enhanced reasoning), or Max (aiming for maximum capability), thereby avoiding unnecessary overhead for simple tasks.
In the Artificial Analysis Coding Agent Index benchmark, Copilot Fusion Max scored 60 points, just two points lower than Claude Code running Opus 5.5 Max. However, its average API costs are roughly half, token consumption is cut in half, and task completion speed is about 40% faster. This feature is now fully available across all GitHub Copilot plans, supporting Visual Studio Code, CLI, applications, and JetBrains IDEs. There are no additional orchestration fees—developers are billed only for actual model calls—and a 30% discount on model charges is offered through November 30, 2026. Enterprise administrators can control which models are available and set them as the default experience. Additionally, the system supports experimental local model routing—for example, invoking Microsoft MAI Code 1.1 Flash on Windows devices.
Microsoft simultaneously introduced Copilot Harness, developed by the GitHub team, which has already empowered multiple products, including Copilot Studio and VS Code, and is now open to developers via Microsoft Foundry. Multi-model intelligent orchestration is emerging as a key evolutionary trend in AI programming tools, promising a better balance between performance and cost.