GitHub Copilot Fusion is generally available
Pick Fusion and an effort level. Copilot routes models for the task. Paid plans get a limited-time 30% discount on those model costs.
Official source (opens in a new tab)
GitHub's changelog for 28 October 2026 says Copilot Fusion is generally available. In the model picker you select Copilot Fusion, then an effort level, instead of a single model. Fusion Standard favors versatile models for everyday coding. Fusion High favors capable models for harder tasks. Fusion Max favors the most capable models. GitHub says Fusion can answer with one model, draft on a cheaper model and escalate, or draft, critique, and revise across models.
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 average API cost per task, about half the tokens, and about 40% less time. That figure is GitHub's. You are billed for each model Fusion uses, at that model's standard rate. For a limited time, paid Copilot plans get a 30% discount on model costs while using Fusion. Administrators can limit which models Fusion may use and can set Fusion as the default. Fusion is on all Copilot plans in Visual Studio Code, the GitHub Copilot app, GitHub Copilot CLI, and JetBrains IDEs.
A second changelog the same day puts local-model routing in experimental preview for the CLI, the Copilot app, and Visual Studio Code. On supported Windows devices, Copilot can download MAI Code 1.1 Flash through Windows ML. The CLI can also route to MAI Code 1.1 Flash or Qwen3.8 Flash Next on a local OpenAI-compatible server on Windows, macOS, and Linux. You select Fusion Standard for Copilot to choose between eligible local and cloud models. GitHub says routing decisions still happen in the cloud, so a local model does not make the session offline. The app's automatic local routing is limited to managed Windows ML models.
Who it's for: someone who already has Copilot and does not want to switch models by hand for each prompt.
Fits: Best for Vibe Coding. GitHub Copilot stays fourth. Fusion changes which model runs. It does not become the in-editor reject-hunk loop that keeps Cursor first. The Copilot alternatives page keeps that boundary. The rank is unchanged.