MikhbarMIKHBAR
Artificial Intelligence

HydraFusion Research Preview Expands to VS Code and GitHub

The HydraFusion research preview is now officially available in Visual Studio Code and the GitHub Copilot app, expanding its multi-model orchestration capabilities beyond the Copilot CLI.

HydraFusion Research Preview Expands to VS Code and GitHub

HydraFusion Expands Beyond Copilot CLI

The GitHub Changelog has announced that the HydraFusion research preview is now available in Visual Studio Code and the GitHub Copilot app, expanding its reach beyond the Copilot CLI. Bringing HydraFusion to additional surfaces was identified as the top request from early user feedback, driving this latest release to broader platforms.

Rather than functioning as a single standalone model, HydraFusion appears directly within the model picker and orchestrates multiple underlying models. By evaluating capability signals for reasoning, code generation, debugging, and tool use, the system automatically selects the most efficient execution pattern required to meet the desired quality bar for any given task.

Understanding Multi-Model Orchestration Workflows

HydraFusion dynamically explores how GitHub Copilot can both select a specific workflow and coordinate multiple models within a single turn. To accomplish this optimization, the system relies on one of three distinct execution workflows depending on the requirements of the prompt.

The first workflow is the Single pattern, where one selected model directly solves the task. The second is the Cascade pattern, in which an efficient model drafts an initial solution and a quality gate determines whether to accept it or escalate the task to a stronger model.

The third execution workflow is the Critique pattern. In this approach, one model drafts a result while an independent read-only critic drawn from a different model family reviews it. This review process follows the same review pattern utilized in Rubber Duck, allowing the drafting model to revise its output once based on the critique.

Enhanced Experience and Workflow Transparency

Alongside its expansion to new surfaces, this release introduces several improvements designed to enhance the overall user experience during complex coding tasks. Users can now enjoy greater transparency into what HydraFusion is doing at each individual step of a workflow.

Furthermore, the system now delivers real-time progress updates much more frequently while it works. These frequent progress reports provide clearer status indications during long tasks, making it significantly easier to determine that HydraFusion is still actively processing even when a request takes additional time to complete.

Configuration and Availability in Visual Studio Code

For developers wishing to use HydraFusion within Visual Studio Code, the feature requires version 1.140 or later, or alternatively VS Code Insiders. Users can select HydraFusion directly from the Copilot Chat model picker.

If the option does not immediately appear, users must enable the relevant configuration. Developers can open Settings, search for 'HydraFusion', and turn it on, or ensure chat.copilot.hydraFusion.enabled is active. For individuals who receive their Copilot access through an organization or enterprise account, an administrator may be required to allow preview features within the organization or enterprise settings.

Access Requirements and Research Context

HydraFusion is currently available to Copilot Pro, Pro+, Business, and Enterprise tier users. For users operating under Copilot Business and Enterprise accounts, administrators must explicitly enable preview features before developers can access the tool.

Because HydraFusion remains categorized as a research preview, it remains subject to change as development continues. Developers interested in diving deeper into the technical architecture can review the official HydraFusion documentation for comprehensive implementation details.

Additionally, readers can examine the background research and benchmark results in Project HydraFusion: Frontier quality via multi-model orchestration, or participate in discussions and share feedback directly with the community through the GitHub Community platform.

Sources

Continue chronologically

Related entity coverage