GitHub made Copilot Fusion generally available on 28 October, according to the GitHub changelog. Fusion is a new entry in Copilot’s model picker. Choose it and Copilot decides which models handle your task, so you stop guessing whether a job needs the expensive one.
The case for it is cost. GitHub says Copilot Fusion Max scored 60 on the Coding Agent Index from Artificial Analysis, two points behind Claude Code running Opus 5.5 Max. On cost, it claims Fusion needed roughly half the API spend per task on average, used around half the tokens, and finished about 40% faster. Those numbers are GitHub’s own reading of the index, and the changelog does not break them down by task type.
You pick an effort level and Fusion handles the rest. Standard favors versatile models for everyday coding, High favors more capable ones for hard problems, and Max reaches for the strongest models on the most demanding work. For each prompt, Fusion then chooses which models to call and how they work together.
It has three ways of working. Single sends the task to one model. Cascade has a cheaper model write a draft, then runs a quality check that either accepts it or escalates to a stronger model. Critique has one model draft, a second review, and the first revise.
Administrators keep a hand on the wheel. Fusion only draws from models already enabled in the enterprise and organization policies. Enterprise admins can see the list Fusion chooses from, set each model’s policy state independently, and switch a model off for Fusion while leaving it on elsewhere. They can also make Fusion the default at any effort level through enterprise managed settings.
The one customer testimonial comes from AT&T. Hari Pokhrel, a principal software engineer there, says Fusion Standard has improved token efficiency across the company. His description of the old habit will sound familiar: a developer picks a powerful model for a hard task and never switches back for routine edits. In his words, “Copilot Fusion eliminates that friction.”
Billing works per model. You pay each model Fusion uses at that model’s standard rate, which means a Critique run is billed for at least two models. For a limited time, people on paid Copilot plans get 30% off model costs when they use Fusion. GitHub has not said when the discount ends.
Fusion is available on every Copilot plan across the VS Code editor, JetBrains IDEs, the Copilot app from GitHub, and the Copilot command-line tool, with more clients promised. Research preview users who turned on auto, HydraFusion, or the Max variant of HydraFusion are moved to the matching Fusion effort levels automatically.
The change matters beyond one product. Model choice has been a daily developer decision, and Fusion hands it to the vendor, along with the line item it controls. Budgets that used to be set by which model a team allowed now depend on a router’s judgment.
The discount has no stated end date, so any cost comparison run during it will flatter Fusion. Price it at standard rates before you make it the default.
Reported by the GitHub changelog on 28 October 2026.
