GitHub Copilot Fusion Is Generally Available: No More Picking Models, It Routes and Orchestrates Claude and GPT per Task, Scoring 2 Points Below Claude Code With Opus 5.5 at About Half the Cost per Task in Third-Party Tests

You pick a dispatcher, not a model

GitHub announced at its Universe conference on October 28 that Copilot Fusion is generally available on all Copilot plans, in VS Code, the GitHub Copilot app, Copilot CLI and JetBrains IDEs. With Fusion selected in the model picker, developers no longer choose a model per task, only an effort level: Standard for everyday coding, High for harder tasks and deeper reasoning, and Max for the most demanding work.

It merges two things GitHub already had: the multi-model orchestration of Project HydraFusion, previously offered as a research preview, and the fine-tuned selector behind Copilot's auto model selection. Users who had enabled auto, HydraFusion or HydraFusion Max are moved to the matching effort level automatically.

Three ways to orchestrate

GitHub's changelog lists three workflows Fusion uses as needed:

• Single: one model solves the task directly;
• Cascade: an efficient model drafts, and a quality check decides whether to accept the draft or escalate to a stronger model;
• Critique: one model drafts, another reviews, and the original model revises.

GitHub has not published a fixed model list. In its testing, Artificial Analysis saw a pool that included Claude Opus 5.5, Claude Sonnet 5.5, GPT-6 Astra and GPT-6 Sol.

Scores and the bill

Artificial Analysis tested Fusion Max in Copilot CLI: 60 on its Coding Agent Index, against 62 for Claude Code with Opus 5.5 Max, at $5.59 per task versus $11.42, using about 7.2 million tokens per task, roughly half as many. By component, it leads Terminal-Bench 4.0 at 71, ranks third on FrontierCode at 51, and is weaker on SWE-Atlas-QnA-AA at 57.

Billing is at the standard rate of every model Fusion invokes, so a single cascade or critique can involve more than one model; users on paid plans get a limited-time 30% discount on model costs. Enterprise administrators can restrict which models Fusion may use, disable a model for Fusion only, or make it the default.

Local routing still needs the network

The same day, GitHub put local-model routing into experimental preview: Fusion Standard can choose between local and cloud models. Supported Windows devices can download and configure MAI Code 1.1 Flash directly through Windows ML; in Copilot CLI, users on Windows, macOS and Linux can also connect MAI Code 1.1 Flash or Qwen3.8 Flash Next through their own local OpenAI-compatible server. GitHub notes that routing decisions still happen in the cloud, so local inference does not make a session offline, and local models may have different safety protections from cloud models.