GitHub Copilot Adds Fusion, New App, and Code Review Max: What's New

GitHub used the opening keynote of GitHub Universe 2026 to ship a wave of concrete Copilot updates, not just roadmap slides. On October 28 at Fort Mason Center in San Francisco, GitHub CEO Jay Parikh walked through a redesigned GitHub Copilot app, a new multi-model routing system called Copilot Fusion, a higher-end code review tier, and a pricing change that folds Copilot into GitHub Enterprise without a separate subscription. Several of the features, including Copilot projects and the new Enterprise terms, are live today; self-serve access to the enterprise pricing model is coming later.

What GitHub Announced at Universe 2026

GitHub's recap post frames the keynote around what it calls an "agentic software delivery system," built on three goals: velocity (getting from idea to shipped software faster and more cheaply), intelligence (context that compounds across a continuous learning loop), and quality (catching problems before they ship). Underneath that framing, GitHub announced a specific set of new Copilot surfaces and capabilities, most of which are already shipping rather than slated for a future date. The company published the details in a post titled "GitHub Universe 2026: Evolving every layer of the stack," written by CEO Jay Parikh and published on GitHub's official blog.

The headline items are a new Copilot app, a multi-model orchestration feature called Copilot Fusion, multi-repo "Copilot projects," cross-device session continuity, smarter Copilot automations, a new top-tier code review mode, and a restructured enterprise pricing model. Each is covered in detail below, using only the specifics GitHub itself has published.

The GitHub Copilot App: One Surface From Issue to Merge

GitHub describes the new Copilot app as "designed from the ground up for agentic software delivery." Rather than being a single coding assistant bolted onto an editor, it's pitched as a hub that connects agents and tools across multiple projects so a developer can direct work "without making every handoff" themselves. GitHub is also positioning it as more than a code-generation tool: the company says you can write specs, run design discussions, and analyze telemetry inside the same agent-first surface.

That positioning matters because it's the container for several of the other announcements below. Copilot projects, session continuity, and the newest automations behavior are all described as living inside this app rather than as standalone features bolted onto VS Code or github.com.

Copilot Fusion: Multi-Model Orchestration

Copilot Fusion is GitHub's answer to the fact that no single model is best at every coding task. According to GitHub, Fusion "evaluates what a task needs across reasoning, code generation, debugging, and tool use, then chooses how to execute across one or more models" — meaning a single request might get routed to, or split across, different underlying models depending on what the task actually requires, rather than requiring the developer to pick a model manually.

GitHub backed the feature with a specific benchmark claim, which is worth quoting directly rather than paraphrasing: in the Artificial Analysis Coding Agent Index, "Fusion Max scored within two points of Opus 5.5 Max at half the cost and 40% less time per task." That's the only benchmark figure GitHub published for Fusion in the keynote recap — there's no separate breakdown by task type, and GitHub did not publish the raw Coding Agent Index scores for either model, so the "two points" gap should be read as GitHub's own characterization rather than an independently verifiable number. It's also notable that GitHub named a specific comparator model, Opus 5.5 Max, rather than speaking generically about "frontier models," which is a shift from how GitHub has talked about model competition in the past.

Fusion arrives on the heels of other recent model changes inside Copilot. GitHub shipped GPT-6.1 Terra as a selectable Copilot model earlier this year, and more recently retired a set of older models, as detailed in our coverage of the Copilot model deprecations that took effect on October 19. Fusion effectively builds on top of that multi-model foundation: instead of developers choosing a model per session, Fusion is meant to make that choice automatically, task by task. For developers weighing GitHub's agent against other AI coding tools, it's also worth reading how Copilot compares with Claude Code, since Fusion's benchmark claim explicitly references an Anthropic-class model as the comparison point.

Copilot Projects: Multi-Repo Agent Sessions

Copilot projects is available natively in the Copilot app starting today. GitHub describes it as a way to "run agent work across multiple repos in fully managed, parallel sessions," with a single place to track progress, make decisions, and direct what happens next. The pitch is that a developer gives Copilot a goal rather than a single ticket, and the agent coordinates work that spans more than one repository without the developer having to manually track each thread.

GitHub frames the benefit as time reallocation: less time spent managing execution across repos, more time spent deciding what's worth building in the first place. The company did not publish a maximum number of concurrent repos or sessions, or details on how conflicts between parallel sessions touching shared dependencies are resolved.

Session Continuity Across Devices

Session continuity lets a developer start and follow the same cloud agent session across the GitHub app, VS Code, the CLI, and the web, rather than being tied to whichever surface they started in. GitHub's own example in the keynote recap: start a session from your phone at standup, steer it from VS Code at your desk, and check in from github.com at lunch, while the agent keeps working in the cloud between those check-ins.

GitHub attributes this to what it calls a "multi-model harness" running underneath, which it describes as powerful, fast, and cost-efficient, without publishing further technical detail on how session state is synchronized across clients.

Copilot Automations Learn From Their Own Results

Copilot automations are AI-powered workflows that live inside a repo and run in the cloud. They can be triggered three ways: manually, in response to an event, or on a recurring schedule — and GitHub points to issue triage and code review as example use cases. The capability itself isn't brand new, but GitHub announced a meaningful extension at Universe: automations can now "learn from their outcomes and propose improvements for you to review and approve." In other words, rather than running the same fixed workflow indefinitely, an automation can surface a suggested change to its own behavior based on how previous runs went, with a human still approving any change before it takes effect.

GitHub frames this as the practical expression of the "continuous learning" idea that ran through the whole keynote: each completed run feeding information back into the system so the next run is better calibrated, rather than every run starting from a blank slate.

Copilot Code Review Max: A Deeper Look at What Ships

As more code ships through agents, GitHub's argument is that review capacity has to scale with it, or the quality bar slips. Copilot code review already shipped in Lite and Balanced tiers for day-to-day reviews; Universe added a third, heavier tier called Copilot code review Max. GitHub says Max "posted the highest F1 score on ReviewBench for high-severity issues," positioning Lite and Balanced as the fast, everyday options and Max as the deeper pass for changes that warrant more scrutiny.

GitHub did not publish the actual F1 score for Max, nor comparative F1 numbers for Lite or Balanced, so the "highest" claim should be read as GitHub's characterization of its own internal ReviewBench results rather than a published, independently auditable figure.

Pricing and Availability for Enterprise Customers

The other major Universe 2026 announcement is a pricing restructure. Starting today, GitHub Enterprise customers can use all Copilot features — including the Copilot app, code review, and automations — without buying a separate Copilot subscription on top of their Enterprise plan. Instead, customers pay for the AI credits they actually use. GitHub says this is available through GitHub Sales today, with self-serve access "coming soon." Existing Copilot Business and Copilot Enterprise customers aren't forced to switch: GitHub says they can keep their current plans or move to the new model when they're ready.

GitHub did not publish specific AI credit pricing, conversion rates, or usage caps in the keynote recap, so organizations evaluating the change will need to talk to GitHub Sales or watch the GitHub Changelog for pricing detail as self-serve rolls out.

How This Fits Into Copilot's 2026 Roadmap

Universe 2026 is the latest step in a string of Copilot changes this year rather than a one-off event. GitHub moved to offer GPT-6.1 Terra as a model option, then trimmed the model lineup in October's deprecations, and is now layering Fusion on top so developers don't have to track which model fits which task themselves. The keynote also tied Copilot's roadmap to infrastructure investment: GitHub says it has added 4 million CPU cores, 160 PB of high-speed storage, and 50 Tb/s of network capacity so far this year, alongside a new Git architecture and an in-progress multi-region migration to Azure that GitHub expects to complete early next year. None of that is Copilot-specific, but GitHub frames it as the capacity underpinning features like Fusion and session continuity, which depend on running agent sessions continuously in the cloud.

GitHub also used the keynote to reiterate that the platform stays open and extensible, specifically naming Cloudflare, Datadog, and Figma as third-party tools teams can keep using alongside Copilot — a signal that GitHub isn't positioning Copilot as a closed replacement for existing toolchains.

Frequently Asked Questions

What to Watch Next

The concrete next milestones to watch, based on what GitHub has published so far: the self-serve rollout of the new Enterprise Copilot pricing, which GitHub has only described as "coming soon"; any published ReviewBench scores or methodology for Copilot code review Max beyond the "highest F1" claim; and further detail on Copilot Fusion's model routing, including whether GitHub publishes the raw Artificial Analysis Coding Agent Index scores behind its "within two points" comparison. GitHub's Changelog and Copilot documentation are the most likely places those details will land first, alongside any follow-up posts from the GitHub Universe event site covering Day 2 sessions.

Written byFelix Moreau

Felix Moreau writes Pandromeda's software coverage and how-to guides. He covers Windows, macOS and Linux updates, the apps people rely on, emulators and developer tools, and turns official documentation into clear, numbered steps that work on the current version.

More from Felix MoreauAll Software & Guides coverage