GitHub Universe 2026 Copilot Overhaul: Fusion Is Generally Available, Projects Are in Preview, and Code Review Goes Deeper

GitHub rebuilt Copilot around agents at GitHub Universe 2026. The GitHub Universe 2026 Copilot overhaul was announced on 28 October 2026, four days before this article was written on 1 November, and it is a platform overhaul rather than a single-model launch: a Copilot app designed for agentic software delivery, Copilot Fusion generally available, Copilot Projects in public preview, and a deeper Copilot Code Review Max tier, also in public preview. GitHub Enterprise customers also gain access to all Copilot features without a separate subscription, with AI-credit costs still applying. The availability states matter: Fusion is generally available, Projects and Code Review Max are previews, and the Enterprise change is live through GitHub Sales, with self-serve still to come.

The Copilot app is now built for finished work, not just code completion

GitHub frames the change around completed work, not more code. The Copilot app is meant to carry work from issue to merge in one place, connecting agents and tools across projects so the developer directs the work without making every handoff. It covers more than producing code: writing specifications, presenting design discussions, and analysing telemetry in the same agent-first surface, with developers setting the direction and staying accountable for what ships while agents implement and verify changes.

github-universe-copilot-1.jpg


Illustration generated for INORYVA — not a real person, product screen, or place.

Two supporting pieces make the app usable across a working day. Session continuity lets a developer follow the same cloud agent session across the GitHub app, VS Code, the command line, and the web, so work does not restart when the tool changes. Copilot automations run AI-powered workflows inside a repository in the cloud, started manually, by an event, or on a schedule, and can now learn from outcomes and propose improvements for a human to approve. GitHub is not the only vendor rebuilding its assistant around background agent work; Anthropic's Claude Duet takes a proactive-agent approach to the same shift from chat to delegated work.

Copilot Fusion is generally available: pick an effort level, not a model

Copilot Fusion removes the model choice from each task. The developer selects Copilot Fusion in the model picker and chooses an effort level — Standard for everyday coding, High for challenging tasks and deeper reasoning, or Max for the most demanding work — and Fusion routes and orchestrates models for each prompt. It combines Project HydraFusion orchestration with Copilot auto model selection, using three patterns: Single (one model solves the task), Cascade (an efficient model drafts, a quality check accepts or escalates), and Critique (one model drafts, another reviews, the original revises).

GitHub says that, in the Artificial Analysis Coding Agent Index, Copilot Fusion Max scored 60, within two points of Claude Code with Opus 5.5 Max, at about half the average API cost per task, using about half as many tokens and taking about 40% less time per task. That is GitHub's own comparison, citing a third-party index — not an independent result.

Fusion is available on all Copilot plans in Visual Studio Code, the GitHub Copilot app, GitHub Copilot CLI, and JetBrains IDEs, with more clients coming soon. There is no separate Fusion fee: users are billed for each model Fusion uses at that model's standard rate, with a limited-time 30% discount on model costs for paid plans. Administrators control which models Fusion can use through model policies and can set Fusion as the default under Enterprise managed settings. Routing a task to the right model differs from answering thousands of small bounded decisions in production software, the job OpenAI built the OpenAI Decisions API for.

Copilot Projects are in public preview: one goal across many repositories

Copilot Projects give multi-repository work one home in the Copilot app. A developer starts a project, describes the goal, and adds context such as repositories, design mocks, Slack threads, or meeting notes, then chooses the models and tools for the job. Copilot runs as many agents as the work needs in fully managed, parallel sessions, and results such as pull requests come back for review — one place to track progress, make decisions, and direct what happens next.

github-universe-copilot-2.jpg


Illustration generated for INORYVA — not a real person, product screen, or place.

That fits work that does not sit in one repository: a feature spanning several repositories, or a launch plan pulling in specifications and designs without touching a repository. Project automations cover repeatable tasks, such as creating a new promotional video whenever a feature lands, while a focused task in a single repository can still start directly from that repository. Projects are in public preview, not general availability, so GitHub has not presented them as finished. The same agent-first direction is visible in community tooling for coding agents, including the Impeccable GitHub skill, which gives agents a design vocabulary for the interfaces they build.

Copilot Code Review Max is a deeper, dearer review — still in public preview

Copilot code review now has three effort levels. Lite gives fast, targeted feedback on bugs, security vulnerabilities, and style inconsistencies. Balanced, the default, uses a higher-reasoning model for longer analysis of complex logic, security-sensitive code, and cross-service changes. Max is built for high-risk changes such as authentication logic: several independent review agents run across models from different providers, and an aggregation agent reconciles their findings, resolves disagreements, and double-checks accuracy into a single review. Max takes longer and uses more AI credits, and GitHub's docs state that Max is in public preview and subject to change.

github-universe-copilot-3.jpg


Illustration generated for INORYVA — not a real person, product screen, or place.

GitHub's own docs give estimated AI-credit consumption per review, and the figures are estimates only. Lite typically ranges from $0.05 to $1 per review, averaging about $0.25; Balanced from $0.25 to $5, averaging about $1.50; Max from $1.50 to $25, averaging about $10. Costs rise with pull-request size and custom instructions, can fall outside the ranges, and exclude GitHub Actions minutes. Administrators can set a maximum review effort level for an enterprise, organisation, or repository, so Max can be kept out of places where its cost is not justified.

The same 28 October changelog also made Copilot approvals generally available. Every review includes an approval assessment stating whether Copilot considers the pull request ready to merge. When approvals are turned on, Copilot can approve the pull request, and that approval counts toward required approvals. Approvals are off by default, administrators turn them on, and an approval is dismissed if new commits are pushed afterwards. Max and approvals are available with Copilot Pro, Pro+, Max, Business, and Enterprise plans.

Enterprise customers no longer need a separate Copilot subscription

Starting at Universe, GitHub Enterprise customers can use all Copilot features without buying a separate subscription, including the Copilot app, code review, automations, and more, and pay for the AI credits they use. The change is a subscription change, not free access: credit and model costs still apply. It is available through GitHub Sales, with self-serve coming soon, and existing Copilot Business and Copilot Enterprise customers can keep their current plans or move over when they are ready.

Key Takeaways

  • GitHub announced the Copilot overhaul at GitHub Universe on 28 October 2026: a Copilot app built for agentic software delivery, Copilot Fusion generally available, Copilot Projects in public preview, and Copilot Code Review Max in public preview.
  • Copilot Fusion replaces the per-task model choice with an effort level — Standard, High, or Max. GitHub says 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 API cost and tokens and 40% less time; that is GitHub's claim, not an independent result.
  • Copilot Projects run agent work across multiple repositories in parallel sessions from one goal, returning results for human review.
  • Code Review Max runs several independent review agents across models from different providers, reconciled into one review. GitHub's docs estimate about $0.25 (Lite), $1.50 (Balanced), and $10 (Max) per review on average, excluding GitHub Actions minutes. Copilot approvals are generally available and off by default.
  • GitHub Enterprise customers can use all Copilot features without a separate subscription and pay for the AI credits they use, through GitHub Sales now, with self-serve coming soon.

Conclusion

The GitHub Universe 2026 announcements change what Copilot is for: less a helper that completes the next line, more a system that carries work from a stated goal to a reviewed pull request, with the developer directing throughout. Test Fusion's effort levels and the Enterprise change first; measure Projects and Code Review Max on high-risk work before making them defaults, especially Max at roughly $10 a review.

Sources

  • GitHub Blog, "GitHub Universe 2026: Evolving every layer of the stack" (Jay Parikh, published 28 Oct 2026): https://github.blog/news-insights/company-news/github-universe-2026-evolving-every-layer-of-the-stack/
  • GitHub Changelog, "Route and orchestrate models with GitHub Copilot Fusion" (published 28 Oct 2026, 09:00 PT; Fusion generally available, benchmark citation, effort levels, billing): https://github.blog/changelog/2026-10-28-route-and-orchestrate-models-with-github-copilot-fusion
  • GitHub Changelog, "GitHub Copilot app, Universe releases" (published 28 Oct 2026, 08:55 PT; Copilot Projects public preview): https://github.blog/changelog/2026-10-28-github-copilot-app-universe-releases
  • GitHub Changelog, "Copilot code review: Merge with confidence with Max and approvals" (published 28 Oct 2026, 08:30 PT; Code Review Max public preview, approvals generally available): https://github.blog/changelog/2026-10-28-copilot-code-review-merge-with-confidence-with-max-and-approvals/
  • GitHub Docs, "About GitHub Copilot code review" (estimated consumption per review by effort level; Max public preview note): https://docs.github.com/en/copilot/concepts/code-review/code-review
  • Artificial Analysis Coding Agent Index — as cited by GitHub in the Copilot Fusion changelog; no independent re-test. No separate URL claimed.




Comments

Post a Comment