Intel / article

GitHub Copilot adds HydraFusion, a multi-model routing option

Source: github.blog
Author: GitHubTop Viber
Date: Published Sep 30, 2026

From the Daily Brief

HydraFusion moves multi-model orchestration from custom harnesses into the editor. In the Copilot app and VS Code it shows up as a single entry in the model picker. Behind that entry, GitHub routes each task across several models, using some to draft, others to critique or revise, and escalating to stronger models when needed, then returns one answer. The appeal is that users get a draft-and-review loop without building it. The tradeoff is visibility: when one selection hides several models, cost, latency and failure attribution become harder to reason about. Teams that adopt it should compare results and usage against a single fixed model on their own repositories before making it the default.

Read the 2026-10-01 Brief →

Context

HydraFusion shows up in Copilot's model picker like any other model, but choosing it hands the task to an orchestrator that decides how several models should work on it. GitHub's September 30 changelog describes three workflows it picks between, using signals about each model's reasoning, code generation, debugging and to find the most efficient pattern that meets a quality bar. In Single, one chosen model does the task. In Cascade, an efficient model drafts and a quality gate either accepts the draft or escalates to a stronger model. In Critique, a read-only critic from a different model family reviews the draft and the drafting model revises once. That is the difference from Copilot's Auto option, which picks one model per request; HydraFusion also chooses the workflow and coordinates multiple models within a turn.

This release moves the research preview beyond Copilot CLI into VS Code 1.140 or later, including Insiders, and the GitHub Copilot app, which GitHub says was the top request from early feedback. It also adds more frequent progress updates and clearer step-by-step status, so long tasks visibly keep working. In VS Code it is selected from the Copilot Chat picker, with a chat.copilot.hydraFusion.enabled setting if it does not appear; in the app it has to be switched on in Settings first. It is open to Copilot Pro, Pro+, Business and Enterprise users, Business and Enterprise administrators must allow preview features, and GitHub says it remains a research preview subject to change.

The changelog makes no performance or cost claims itself. It points to a separate research write-up for results, which was not part of the evidence read for this explanation, so how much quality or cost HydraFusion gains over picking a single model is not established here.

Terms in this piece · Glossary

tool use — A model's ability to call external functions — run code, search the web, edit files — instead of only generating text.
benchmark — A standard public test set for comparing AI models — the shared scoreboards behind every "model X beats model Y" claim.

Read the source github.blog