GitHub announced GitHub Copilot Fusion at GitHub Universe on October 28, 2026. The system selects and combines models from different labs according to a development task's needs. Fusion is generally available on all Copilot plans in Visual Studio Code, the GitHub Copilot app, GitHub Copilot CLI and JetBrains IDEs. The accompanying code review Max release is in public preview.
Their different roles matter from the start. Fusion manages model selection while a task is performed; Max is a review effort level for checking code before integration. The names do not denote one pricing package or identical availability. GitHub's announcements also distinguish Fusion billing, review costs and the new GitHub Enterprise access arrangement.
Choose effort rather than a model
According to GitHub, Fusion combines Project HydraFusion's orchestration with Copilot's automatic model selection. Users select it in the model picker and choose an effort level. The system decides which models will handle the task and how they work together, based on reasoning, code generation, debugging and tool-use requirements.
Fusion Standard targets everyday coding. Fusion High prioritizes capable models for challenging tasks and deeper reasoning, while Fusion Max targets the most demanding work. These are system effort levels, not newly trained GitHub models. Developers still make a practical choice about the balance between depth and resources.
GitHub describes three execution patterns. Single uses one model to solve the task. Cascade starts with an efficient model, then uses a quality check to accept its response or escalate to a stronger model. In Critique, one model drafts, another reviews and the original revises. Selecting Fusion therefore does not mean every request always invokes several models.
What the savings claim actually measures
GitHub reports that Fusion Max scored 60 on the Artificial Analysis Coding Agent Index, within 2 points of Claude Code with Opus 5.5 Max. It says this used about half the average API cost per task, about half as many tokens and about 40% less time per task. This is a benchmark comparison, not a promise of identical savings in every repository.
Billing is more straightforward than the performance comparison: users pay for each model Fusion uses at that model's standard rate. For a limited time, paid Copilot plans receive a 30% discount on model costs when using Fusion. The launch announcement gives no end date, and this is not presented as a permanent discount on an entire Copilot bill.
There is consequently no single announced Fusion price per task. Work involving several models can produce several billing components, with cost depending on routing decisions and consumption. Teams should distinguish model rates, the temporary discount and GitHub's benchmark savings. These describe different things even when presented together.
code review Max goes deeper
code review Max joins Lite and Balanced. GitHub says it runs independent review agents using models from different providers, then an aggregation agent reconciles findings, resolves disagreements and checks accuracy. It targets sensitive changes, such as authentication logic or a shared library used by several services, where a fault can affect more than the edited file.
GitHub says Max achieved the highest F1 score on ReviewBench for high-severity issues. F1 combines precision and recall; it does not promise that every bug will be caught. The company explicitly says Max takes longer and consumes more AI credits than Lite or Balanced. The announcement provides no fixed price for an individual Max review.
Max is in public preview on Copilot Pro, Pro+, Max, Business and Enterprise plans. Administrators can cap review effort. Separately, Copilot approvals are generally available: when enabled, a Copilot approval can count toward a repository's required approvals. They are disabled by default, and an existing approval is dismissed when new commits are pushed afterward.
Enterprise access changes, usage still costs money
GitHub also announced that Enterprise customers working with a GitHub or Microsoft account team can obtain Copilot access for every Enterprise-licensed user without a separate Copilot subscription. Access includes Copilot Enterprise features and models, but no monthly AI credit allocation per seat. Usage is metered against the enterprise budget from the first request.
Users in the enterprise account without a GitHub Enterprise license can receive Copilot Standalone for 10 dollars per user per month, also without monthly AI credits. Self-service availability will arrive in stages. Copilot Business and Copilot Enterprise remain available, and existing self-service subscriptions are unaffected. Removing a separate subscription does not make computation free.
Model selection also remains subject to organizational policy. GitHub says Fusion uses only permitted models, and administrators can make it the default at their preferred effort level. Automatic routing therefore does not replace the organization's decision about which providers and models may participate in the first place.
Octoverse explains why selection matters
GitHub's accompanying Octoverse report describes the scale of the change. In September 2026, more than 40% of pull requests began with an agent's first draft, compared with 17% in March. That amounted to 110 million in September alone. The figure concerns how a proposed change began, not whether all its code was written or approved without human involvement.
The report also says 61% of Copilot users who selected models themselves used more than one during September. This is a defined population, not all users of every GitHub product. Fusion transfers some of those choices to the system, while Max broadens downstream review. For development teams, both changes connect model selection to more concrete questions: fix quality, review time and the cost of completed work.