GitHub made Copilot Fusion generally available on October 28, 2026. Instead of choosing a model for each task, developers choose Fusion and an effort level, and Copilot decides which models to use for each prompt and how they work together, according to the GitHub changelog.
There are three effort levels. Fusion Standard prioritises versatile models for everyday coding, Fusion High prioritises capable models for harder tasks and Fusion Max prioritises the most capable models for the most demanding work. Fusion runs each task in one of three patterns. Single: one model solves the task directly. Cascade: an efficient model drafts a response, and a quality check either accepts it or escalates to a stronger model. Critique: one model drafts, another reviews and the first revises.
GitHub's performance claim: on the Artificial Analysis Coding Agent Index, Fusion Max scored 60, within two points of Claude Code with Opus 5.5 Max, at about half the average API cost per task, with about half as many tokens and about 40% less time per task.
Billing follows the routing. Users pay for each model Fusion calls, at that model's standard rate, so the cost of a task depends on how it is routed. For a limited time, paid Copilot plans get a 30% discount on model costs when using Fusion. Administrators decide which models Fusion may use, can disable a model for Fusion without disabling it for everyone and can set Fusion as the default model. An AT&T engineer quoted by GitHub says the Standard level improved token efficiency because developers no longer forget to switch back from expensive models.
Fusion is available on all Copilot plans in Visual Studio Code, the GitHub Copilot app, Copilot CLI and JetBrains IDEs, with more clients to follow.
Why this matters for engineering teams
Model choice moves from the developer to a router and an admin policy. That can lower cost, as GitHub's figures suggest, but it also makes cost per task less predictable, because the same prompt may be handled by one model or by three. Teams that adopt it should track spend per task type for a few weeks before making Fusion the default, and keep the list of allowed models aligned with their data policies.
This article is not legal advice.
