Michael JohnsonFor a complex task in Copilot, you can now choose HydraFusion Max: the system determines which workflow and models to use for the solution and review stages. GitHub announced the update on October 14, 2026; HydraFusion and Max are available in research preview in VS Code, the GitHub Copilot app, and Copilot CLI.
HydraFusion acts as an orchestrator for multiple models, rather than as a standalone model. For each request, it selects one of several workflows: one model solves the task; Cascade sends a draft to a judge model, which can accept it or request an enhanced pass; Critique gathers parallel feedback from several models and returns a revised answer. In the update, GitHub said Cascade gained a draft evaluation stage, while Critique makes more active use of tools to inspect the workspace. The HydraFusion update announcement also describes reasoning-effort settings: medium for the initial draft and high for review.
Max is designed for tasks that require deeper review
HydraFusion Max uses the same architecture and orchestration workflows as the standard version, but is intended for more demanding tasks and applies greater reasoning effort where thoroughness is needed. GitHub cites complex agentic tasks as an example; the company describes routine day-to-day work as a fit for automatic selection of a single model.
This gives developers a practical choice between speed and additional review stages. For a small fix, extra passes may add latency and cost without a useful result. For a multi-file change where consistency across edits matters, a separate critique stage may be more appropriate. This is a working hypothesis to test on your own tasks, not a quality improvement guarantee published by GitHub.
Preview calls for monitoring costs and working-copy changes
Access depends on your Copilot plan, enabled experimental features, and organization policies. GitHub’s documentation says HydraFusion selects available models automatically, and each individual pass is charged at that model’s standard rate. As a result, a task using Cascade or Critique may consume more AI credits than a single model call. The HydraFusion documentation describes these conditions, how to view usage, and the current research preview limitations.
Teams have a straightforward way to assess whether orchestration is worth enabling: take a small sample of real tasks, keep the prompts and repository starting points consistent, then compare HydraFusion, Max, and the standard mode. For each result, it is useful to record elapsed time, AI credits, the number of fixes after review, and errors found by tests or a reviewer. This measurement answers the practical question of whether an extra pass improves your particular development process.
Before merging changes, check the diff and run your usual tests: the documentation warns that file changes made during a discarded draft are not automatically reverted. Since these features remain in preview and the models used may change, repeat the evaluation after significant updates. For short everyday requests, the standard model selection is a sensible starting point; HydraFusion Max is worth testing where additional review could save time correcting errors.