GitHub just took the hardest decision in AI coding off your plate: which model to use. At its Universe conference in San Francisco on October 28, the company made Copilot Fusion generally available — a system that picks and combines AI models for each coding task, so developers only choose how much effort a job deserves. If you have ever stared at a model picker wondering whether to burn credits on the flagship model or risk the cheap one, Fusion is built for you.
What Copilot Fusion actually is
Fusion is not a new AI model. It is a routing and orchestration layer that sits inside Copilot's model picker. Pick Fusion, and instead of one model, you get a system: it looks at what a task needs — reasoning, code generation, debugging, tool use — then decides which models should handle it and how they should work together. It combines two things GitHub already built: Copilot's automatic model selection and Project HydraFusion, a multi-model orchestration experiment that ran as a research preview since September. The idea is simple: routine work should never land on the most expensive model by default, and hard work should get more than one brain when it helps.
Three effort levels, zero model-picking
The only choice left to you is effort. Fusion Standard handles everyday coding with versatile models. Fusion High takes on harder, multi-file jobs that need deeper reasoning. Fusion Max throws the strongest models at the most demanding tasks. Under the hood, Fusion runs three patterns: Single, where one model solves the task directly; Cascade, where a cheaper model drafts and a quality check decides whether a stronger model should take over; and Critique, where models check each other's work. You never see any of this — you just set the effort and watch the result.
The numbers GitHub is bragging about
GitHub's own figures, from the Artificial Analysis Coding Agent Index, put Fusion Max within two points of Claude Code running Anthropic's Opus 5.5 Max — while using about half the average API cost per task, about half the tokens, and taking roughly 40% less time per task. Treat those as GitHub's claims, not independent results: no outside testing has verified them yet. Still, the direction is hard to ignore. The coding-assistant race is shifting from "which model is smartest" to "which system spends your tokens wisely." Price and availability
Fusion is generally available now on all Copilot plans — in Visual Studio Code, the
GitHub Copilot app, the Copilot command-line tool, and JetBrains IDEs, with more clients to follow. There is no extra orchestration fee: you pay each model's standard rate for the models Fusion uses. Paid plans get a 30% discount on Fusion model costs through November 30, 2026. The same Universe event also brought the standalone GitHub Copilot app, Copilot Projects for running agent work across repositories, and a change letting GitHub Enterprise customers use every Copilot feature while paying only for the AI credits they consume.
The take
Fusion's real bet is that developers should stop caring which model writes their code. Rivals like Cursor and Windsurf still ask you to think about models, credits, and quotas — GitHub wants the model choice to disappear entirely. If Fusion's cost claims hold up in real projects, "just pick an effort level" could become the default way coding assistants work. The risk is transparency: when a system silently routes your work across models, you need a bill that shows what it chose and why. Watch that part — but as a direction of travel, this is the most developer-friendly AI launch of the week.
Comments
Post a Comment