At its Universe 2026 conference on October 28, GitHub rolled out a batch of new Copilot products: the GitHub Copilot app, built from scratch for agents; Copilot Fusion, an orchestration layer that picks models automatically; plus Copilot projects for running work across multiple repositories in parallel, cloud-based scheduled automations, and a deeper code review mode. Billing for enterprise customers is changing along with them.
The announcement post is signed by Jay Parikh, and its opening line sets the tone:
"Completed work, not more code, is the output that matters."
Fusion: let the system choose the model
Fusion is a multi-model orchestration layer. Depending on what a task demands in reasoning, code generation, debugging and tool calling, it decides whether to hand the job to a single model or to pass it between several. GitHub's results come from the Artificial Analysis coding agent index: Fusion Max scores within 2 points of Opus 5.5 Max, at about half the cost and with 40% less time per task.
These figures were published by GitHub itself. The post does not say which models Fusion actually uses during routing, or in what proportions. A documentation link is live, but GitHub does not state whether this is a general-availability release or a preview.
The Copilot app and projects
The Copilot app is meant to carry work from an issue all the way to a merge, with spec writing, design discussions and telemetry review all handled in one interface. Launching alongside it, Copilot projects can run managed parallel sessions across several repositories, with progress and pending decisions shown in one place. It has been available in the app since October 28.
A single cloud agent session can now move between the GitHub website, VS Code, the command line and the app. Automations can be triggered manually, by events or on a schedule, and a new capability proposes improvements based on run results, which take effect only after a human reviews them. Code review gains a Max mode, which GitHub says achieved the highest F1 score on high-severity issues in ReviewBench; the Lite and Balanced tiers remain for everyday quick reviews.
Enterprise moves to credit-based billing
GitHub Enterprise customers no longer need to buy a separate Copilot subscription. The app, code review, automations and other features are all available, and customers pay for the AI credits they consume. For now, access goes through sales, with self-serve signup listed as "coming soon"; existing Copilot Business and Enterprise customers can stay on their current plans or migrate at any time. The announcement does not mention any change to individual-plan pricing.
On infrastructure, GitHub says it added 4 million CPU cores, 160 PB of high-speed storage and 50 Tb/s of network capacity this year, and expects the multi-region migration to Azure to finish in early next year.
How it compares with similar products
Parallel multi-agent work is a crowded path: Cursor launched Projects a few months ago, and Anthropic's Claude Code is also pushing background agents. GitHub's difference is that issues, pull requests and CI already live on its platform, so an agent can go from picking up a task to submitting a merge without crossing systems. Fusion, meanwhile, takes the choice of model out of the user's hands. For enterprise customers billed by credits, that matters in practice: for the same job, a router that costs half as much shows up directly on the invoice.
Teams in China that use GitHub Enterprise should note that no public price table yet shows the unit price after the billing switch. Before migrating, it is safer to ask sales for the credit conversion rates.
Sources: GitHub official blog Universe 2026 announcement post (used to verify the Fusion score gap, the cost and time ratios, and the infrastructure expansion figures), CocoLoop, Artificial Analysis coding agent index.