GitHub has updated HydraFusion, the Copilot option that can assign drafting and review to different AI models. It now uses medium reasoning effort for drafts and high effort for reviews, giving the checking stage more room to work through a problem. The new HydraFusion Max option also makes extra high reasoning effort available for demanding tasks.

Both remain in research preview in VS Code, the GitHub Copilot app, and Copilot’s command-line tool. GitHub lists access on Pro, Pro+, Business, and Enterprise plans with experimental or preview features enabled, subject to organization policies.

Review can inspect project files directly

HydraFusion chooses how to handle each request and can leave a straightforward task with a single model. Its Cascade workflow can now judge an initial solution and request a repair when needed. Critique uses more tools to inspect the project’s files directly, with multiple models reviewing in parallel, rather than relying as heavily on text summaries.

The model pool now includes Claude Opus 5.5, GPT-6 Astra, GPT-6.1 Sol, and GPT-6 Luna. That does not mean every task uses all four.

Unlike Copilot Auto’s model selection, HydraFusion can choose a sequence of model calls for one task. You hand it the work without assigning each model a role yourself.

Extra passes cost credits, and discarded edits can remain

GitHub’s documentation says each model used is billed at its standard rate. There is no separate HydraFusion charge, but Auto’s discount does not apply. A task using several models can consume more AI credits, Copilot’s usage balance, than a task using one. Review passes also take longer.

There is a separate caution for code changes: discarding a draft does not automatically undo file edits it already made. Review the changed files before saving them in a Git commit, even if the final response looks correct.

If you evaluate the preview on a bug fix, check whether the fix passes your tests, inspect all file changes, and compare total credits consumed. Copilot’s terminal tool shows session usage through /usage. GitHub says HydraFusion is not intended for production workloads during this preview.

End of article