---
title: GitHub Copilot merges routing and orchestration into Fusion mode
url: https://www.elseif.net/github-copilot-merges-routing-and-orchestration-into-fusion-mode
published: 2026-11-02T17:03:33+00:00
language: en
section: Harnesses
source: https://www.silicon.fr/data-ia-1372/github-copilot-routeur-orchestrateur-230036
organizations: GitHub, Michelin
publisher: elseif
---

# GitHub Copilot merges routing and orchestration into Fusion mode

GitHub has integrated its routing and orchestration capabilities into a new mode called Fusion. This development follows the introduction of the Auto mode earlier in the year and the experimental launch of HydraFusion in early September. The Auto mode, which has existed in two versions, uses a ModernBERT encoder to direct requests based on model availability and capabilities. The second version, known as HyDRA, employs a classifier trained to evaluate needs across four dimensions: reasoning, code generation, debugging, and tool usage. It also incorporates operational constraints such as session continuity and dynamic failure filtering. The orchestration component, initially tested as HydraFusion, aims to select the optimal workflow for each prompt rather than the optimal model.

This workflow selection involves three distinct options. The convergence of these two components creates the Fusion mode, which operates at four effort levels: Standard, High, Max, and Fast. The Fast level is not adjustable by users and is reserved for inline chat in VS Code. The Standard level is accessible on all Copilot plans. The High and Max levels are available only on Pro+, Max, Business, and Enterprise subscriptions. While the Auto mode offered a 10 percent discount on model prices, GitHub proposes a 30 percent discount for Fusion, but only for a limited period until the end of November. Administrators can choose to enable Fusion by default in the selector and control the model pool and default effort level for new conversations. The integration of local models remains in preview, currently covering two models.

Early feedback from Michelin, which tested HydraFusion in mid-September, was mixed. The company described the orchestrator as a quality mechanism rather than a substitute for Auto mode, noting a significantly higher cost. User feedback echoed these concerns, stating that HydraFusion tends to choose overly performant models and does not utilize economic sub-agents unless explicitly requested. Frustrations also included a lack of feedback during task execution, as GitHub chose not to display intermediate step results to avoid them being perceived as finished work. Suggestions were made to expose structured telemetry, such as the chosen execution strategy and model families used, instead of the chain of thought.