Summary
On October 14, 2026, GitHub updated HydraFusion, a Copilot research-preview option that orchestrates multiple models into a single response, and introduced HydraFusion Max for harder tasks. The model pool now includes Claude Opus 5.5, GPT-6 Astra, GPT-6.1 Sol, and GPT-6 Luna, with per-step reasoning effort and a new judging-and-repair cascade.
What changed
HydraFusion's model pool now spans Claude Opus 5.5, GPT-6 Astra, GPT-6.1 Sol, and GPT-6 Luna. It applies per-step reasoning effort (medium for drafting, high for review) instead of each model's default, adds a cascade step that judges initial work and requests a repair only when needed, and strengthens critique by relying more on tools that inspect the actual workspace and on parallel critic models. A new HydraFusion Max option uses the same architecture tuned for the most demanding tasks with higher reasoning effort. Both are in research preview in VS Code, the GitHub Copilot app, and the Copilot CLI.
Why it matters
HydraFusion pushes Copilot from single-model selection toward automated multi-model orchestration, removing the per-task model-picking decision and blending frontier models from different providers behind one response. The judging-and-repair cascade and workspace-inspecting critics target reliability on hard coding tasks, and bundling rival frontier models (Anthropic and OpenAI) signals that routing and verification, not any single model, are becoming the differentiator for coding assistants.
Evidence excerpt
HydraFusion orchestrates multiple models to produce a single response. The pool now includes Claude Opus 5.5, GPT-6 Astra, GPT-6.1 Sol, and GPT-6 Luna; a new HydraFusion Max option targets the most demanding tasks with higher reasoning effort.