GitHub Copilot adds HydraFusion Max for harder coding tasks
The research-preview feature that blends several models into one answer now has a frontier-tuned tier, plus a judging step and parallel critics in its workflows.
GitHub updated HydraFusion, the research-preview Copilot feature that orchestrates multiple models into a single response instead of making you pick one, on 14 October. Its model pool now includes Claude Opus 5.5, GPT-6 Astra, GPT-6.1 Sol and GPT-6 Luna, and a new HydraFusion Max option uses the same architecture but is tuned for frontier capability on harder tasks.
The update also changes how reasoning effort is allocated: rather than using each model’s default effort throughout, HydraFusion now applies medium effort for initial drafts and high effort for review. Its Cascade workflow, which escalates from a weaker to a stronger model when needed, gains a judging step that only requests a repair when one is actually required. The Critique workflow moves away from relying on text summaries and now inspects the workspace directly with parallel critic models. Both HydraFusion and HydraFusion Max remain in research preview, available in VS Code, the Copilot app and Copilot CLI on Pro, Pro+, Business and Enterprise plans.
Why it matters: picking the right model for a coding task is still a manual judgement call for most people. HydraFusion is GitHub’s bet that orchestrating several models per request, with cheaper reasoning for drafts and heavier reasoning for review, beats asking one model to do both cheaply and well.
The caveat: it’s still a research preview, so expect it to change, and running multiple models per request is not the cheap option.