GitHub makes Copilot Fusion generally available in Universe 2026 – Unite.AI

GitHub made Copilot Fusion, its multi-model routing and orchestration system, generally available on October 28, 2026, introducing it at GitHub Universe 2026 along with the new GitHub Copilot app, Copilot projects, and a deeper Copilot code review layer.

The announcements came in a Universe 2026 keynote post from Jay Parikh, executive vice president of CoreAI at Microsoft, whose teams include GitHub, and in a series of changelog entries from the same day detailing availability, billing, and preview terms.

How Copilot Fusion routes and orchestrates models

According to the Copilot Fusion changelog, selecting Copilot Fusion in the template selector chooses a routing and orchestration system rather than a specific template. For each request, Fusion determines which models to use and how they work together, combining the multi-model orchestration of the HydraFusion project with the optimized model selection behind Copilot’s automatic model selection feature.

Users choose one of three effort levels. Fusion Standard prioritizes versatile models for everyday coding, Fusion High focuses on demanding tasks and deeper reasoning, and Fusion Max reserves the most capable models for more demanding tasks and complex reasoning. For each task, Fusion executes one of three execution workflows: Single, in which a model directly solves the task; Cascade, where an efficient model processes a response and QA accepts it or switches to a stronger model; and Critique, in which one model drafts, a second model revises it, and the original model revises it.

GitHub said Copilot Fusion Max scored 60 on the Artificial Analysis Coding Agent Index, within two points of Claude Code with Opus 5.5 Max, at about half the average API cost per task, using about half the tokens and about 40% less time per task.

Administrators control which templates Fusion can draw on through corporate and organizational template policies and can disable templates for Fusion without disabling them for everyone. Company-managed settings can also set Fusion, at any engagement level, as the default template for Copilot users.

Hari Pokhrel, a top software engineer at AT&T, said in the changelog that Fusion Standard has improved token efficiency at AT&T by automatically selecting the right model for each task, reducing unnecessary expenses and preserving capacity for higher-value work, and that its decision making has improved over time.

Fusion is billed per template at each template’s standard rate, and users on paid Copilot plans receive a limited-time 30% discount on template costs when they use it. The system is available on all Copilot plans in Visual Studio Code, the Copilot GitHub app, the Copilot GitHub CLI, and JetBrains IDEs, with additional client support to come. Developers who have enabled Auto, HydraFusion, or HydraFusion Max mode during search preview will automatically switch to the corresponding Fusion effort levels.

The Copilot GitHub app and Copilot projects

The GitHub Copilot app is designed for deploying software via agents, Parikh wrote, connecting agents and tools across projects and supporting work beyond code, including writing specs, presenting design discussions, and analyzing telemetry in a single surface.

Copilot projects, available natively in the app starting October 28, 2026, allow users to run agent work across multiple repositories in fully managed parallel sessions. The Copilot app changelog labels Copilot projects as a public preview: users describe a goal, attach context such as repositories, design mocks, or Slack threads, choose templates and tools, and receive results as pull requests for review. Project automations handle repeatable tasks, such as creating a promotional video every time a feature arrives.

The app can also use context from connected tools like Slack and Figma, and can generate and edit images from natural language in chat. A computer usage feature, available in public preview on macOS and Windows, allows Copilot to read app content, click controls, insert or edit text, and navigate between desktop apps, including those without API, CLI, or MCP integration. Copilot requires approval before checking an app, and organizations can disable the feature through managed settings.

Session continuity allows users to start and follow the same cloud agent session in the GitHub app, VS Code, CLI, and web.

Code Review Max, Automations and Enterprise Access

GitHub also introduced the Copilot Max code review, which Parikh said posted the highest F1 score on ReviewBench for high severity issues. The existing Lite and Balanced levels remain for daily reviews, with Max earmarked for changes that require a more in-depth look.

Copilot automations create AI-powered workflows within a repository running in the cloud, triggered manually, by events, or on a schedule, for tasks such as issue triage and code review. GitHub said that automations can now learn from their results and propose improvements that users can review and approve.

Starting October 28, 2026, GitHub Enterprise customers can use all Copilot features, including the Copilot app, code review, and automations, without purchasing a separate subscription, instead paying for the AI ​​credits they use. The change is available immediately via GitHub Sales with self-service coming soon; Existing Copilot Business and Copilot Enterprise customers can keep their current plans or change them when they’re ready.

The road from HydraFusion and infrastructure additions

Fusion’s general availability follows the HydraFusion GitHub project search preview introduced on September 4, 2026, after launching its car model selection earlier in the year. In offline evaluations reported with the preview, HydraFusion’s fixed policies improved the quality of verified tasks by 4.9 percentage points at an estimated cost 67% lower than Claude Opus 5 on TerminalBench 2.1, came in 1.5 percentage points at a cost 36% lower on DeepSWE, and by 0.1 percentage points at a cost 65% lower on CheckpointBench, GitHub’s internal benchmark built from Copilot sessions real. The three HydraFusion execution models: Single, Cascade, and Critical are inserted into the Fusion workflow set.

GitHub said it has added 4 million CPU cores, 160 PB of high-speed storage, and 50 Tb/s of networking capacity so far in 2026. Three principles guide its new Git architecture: simplify primitives, decouple storage from compute, and minimize coordination. The company is migrating to a multi-region deployment on Azure and expects to complete that migration in early 2027.