GitHub just made a compelling case that the best AI coding assistant isn’t a single model. It’s a traffic cop that knows which model to call and when.
The company announced Project HydraFusion on September 4, 2026, as a research preview inside GitHub Copilot. The system is a multi-model orchestration layer that dynamically selects from different AI execution patterns to optimize for quality, cost, and speed.
How HydraFusion actually works
At its core, HydraFusion operates across three execution patterns: Single, Cascade, and Critique. The Single pattern routes a task to one model. The Cascade pattern chains multiple models together, escalating complexity as needed. The Critique pattern adds an isolated review step where a separate model evaluates another’s output before shipping it.
This builds on GitHub’s previous Auto model selection feature, which let Copilot pick a model for users. The difference is that Auto made static selections. HydraFusion is dynamic, adjusting its approach mid-task if the situation warrants it.
