GitHub Universe 2026: Copilot Fusion Picks Models for You, and Copilot Is Included in GitHub Enterprise
At its Universe conference, GitHub released Copilot Fusion, a mode that selects and combines models based on the task, added multi-repository projects and computer use to the Copilot app, made automations shareable and capable of suggesting their own improvements, and bundled all Copilot features into GitHub Enterprise with usage-based billing.
GitHub used its Universe conference on October 28 to outline its vision of software development steered by “fleets of agents” orchestrated by developers. Beyond the keynote rhetoric, four concrete announcements alter how Copilot is used: a mode that picks models on behalf of the user, an expanded desktop app, automations that learn from their outcomes, and a new commercial model for large enterprises.
Copilot Fusion: No More Picking Models
Copilot Fusion is now available across all Copilot plans in Visual Studio Code, the GitHub Copilot app, Copilot CLI, and JetBrains IDEs. Rather than selecting a specific model, users select “Copilot Fusion” from the picker, then choose an effort level: Standard for everyday coding, High for difficult tasks, and Max for the most demanding workloads.
Fusion then decides, for each prompt, which models to deploy and how to combine them. Three execution patterns are possible: a single model handles the task; a cascade, where a cost-effective model drafts an answer that an evaluation step either accepts or hands off to a more capable model; or a critique, where one model writes, another reviews, and the first applies corrections. Fusion brings together two existing components: the HydraFusion orchestration project and Copilot’s automatic model selection.
On pricing, each model used is billed at its standard rate, with a 30% discount offered for a limited time to paid subscribers. GitHub claims that Fusion Max scores 60 on Artificial Analysis’s Coding Agent Index, trailing Claude Code with Opus 5.5 Max by just two points, while costing about half as much and running roughly 40% faster per task. This benchmark was published by GitHub and has not been independently verified.
The Copilot App Manages Entire Projects
The GitHub Copilot app gains projects, now in public preview: developers describe an objective, attach relevant context (repositories, mockups, Slack threads, meeting notes), and Copilot spins up as many agents as needed across multiple repositories in parallel before surfacing the results—such as pull requests—for review. A session continuity feature lets developers monitor the same cloud agent session seamlessly from the app, VS Code, the terminal, or the web interface.
The app also introduces computer use, in public preview on macOS and Windows: Copilot can inspect desktop application contents, click, enter text, and switch between software, including tools lacking an API or MCP integration. Copilot prompts for user permission before taking control, and organizations can disable the feature entirely.
Shared Automations That Propose Their Own Fixes
Copilot Automations are cloud-run workflows triggered manually, by an event (an opened issue, a ready pull request), or on a schedule. Two major updates: they can now be shared at the repository level (visible to the team, billed to the repository owner, with actions attributed to Copilot rather than an individual), and they can learn from their outcomes. For instance, an issue-triage automation observes whether human contributors correct its assigned labels, and then proposes an update to its own prompt, which the user must approve or reject before it takes effect.
Copilot Bundled into GitHub Enterprise
The final major announcement targets enterprise customers: since October 28, GitHub Enterprise clients can access all Copilot features without a separate subscription (the app, code review, automations, etc.), paying strictly for the AI credits consumed. The offering is currently routed through GitHub’s sales team, with self-serve purchasing promised “soon,” while existing Copilot Business or Enterprise customers can keep their current plans. GitHub also unveiled Copilot code review Max, a more thorough review tier that it claims leads ReviewBench for high-severity defects.
Our take
The common thread across these announcements is consumption-based pricing. With Fusion, GitHub takes over model selection, simplifying the developer experience while making the actual cost of a task harder to predict; the temporary 30% discount is clearly meant to ease that friction. Bundling Copilot into GitHub Enterprise in exchange for consumed credits points in the same direction: fewer per-seat licenses, more pay-per-task billing, a model well-suited for autonomous agents running in the background. Automations that suggest their own improvements are the most compelling concept here, provided human oversight remains genuine and does not degrade into a rubber-stamp click.
Sources
- GitHub — GitHub Universe 2026: Evolving every layer of the stack Primary source
- GitHub Changelog — Route and orchestrate models with GitHub Copilot Fusion Primary source
- GitHub Changelog — GitHub Copilot app, Universe releases Primary source
- GitHub Changelog — Build shared workflows for your team with Copilot Automations Primary source
- GitHub Changelog — Copilot Automations now learn from outcomes and self-improve Primary source
A primary source comes directly from the company, lab or organization concerned. A secondary source reports the information without being its author: we flag it as such. Spotted an error? See our corrections policy.
