GitHub Copilot Fusion goes GA, routing tasks across models at about half Opus 5.5's cost

What GitHub launched

GitHub announced on 28 October 2026 that Copilot Fusion is generally available in GitHub Copilot. Fusion replaces the model picker choice with a routing system. For each prompt it decides which models to use and how they work together, and the developer only picks an effort level: Standard for everyday coding, High for harder tasks and Max for the most demanding ones.

Fusion merges GitHub's earlier Project HydraFusion research preview with Copilot's auto model selection. It uses three patterns. In Single, one model answers directly. In Cascade, a cheaper model drafts and a quality check decides whether to escalate to a stronger one. In Critique, one model drafts, a second reviews and the first revises. Research preview users on auto or HydraFusion are moved to the matching Fusion levels automatically.

Benchmark and cost claims

GitHub says Fusion Max scored 60 on the Artificial Analysis Coding Agent Index, within two points of Claude Code running Claude Opus 5.5 at max effort. It says Fusion did this at about half the API cost per task, with about half the tokens and about 40% less time per task. Artificial Analysis published its own numbers: $5.59 per task for Fusion against $11.42 for Claude Code with Opus 5.5, and 7.2 million tokens per task, roughly half as many.

Artificial Analysis says Fusion chooses among Anthropic and OpenAI models, including Claude Opus 5.5, Claude Sonnet 5.5, GPT-6 Astra and GPT-6 Sol. It led Terminal-Bench 4.0 with 71 but ranked third on FrontierCode at 51 and lagged on SWE-Atlas-QnA at 57.

Pricing, admin controls and local models

Fusion bills each model it uses at that model's standard rate. For a limited time, paid Copilot plans get a 30% discount on model costs when using Fusion. It is available on all Copilot plans in VS Code, the GitHub Copilot app, Copilot CLI and JetBrains IDEs. Fusion only uses models enabled in enterprise and organization policies, and admins can exclude models from Fusion without disabling them for everyone or make Fusion the default.

In a separate experimental preview, Fusion Standard can also route between cloud models and on-device models. On Windows, Copilot can download Microsoft's MAI Code 1.1 Flash through Windows ML. On Windows, macOS and Linux, Copilot CLI users can point it at MAI Code 1.1 Flash or Qwen3.8 Flash Next on a local server. GitHub notes the routing decision still happens in the cloud.

What to watch next

Routing hides the model choice, so the main question is cost predictability once the launch discount ends. GitHub has not given an end date for the 30% discount. Agent features from the same event are covered in our Agentic Workflows story. More coding tool news is on the AI Tools hub.

Questions people ask

How much does GitHub Copilot Fusion cost?

There is no separate fee. Each model Fusion uses is billed at its standard rate, and paid Copilot plans get a 30% discount on model costs for a limited time.

Which models does Copilot Fusion use?

Only models enabled by your organization's policies. Artificial Analysis says it chooses among Anthropic and OpenAI models such as Claude Opus 5.5, Claude Sonnet 5.5, GPT-6 Astra and GPT-6 Sol.

Can Copilot Fusion use local models?

In an experimental preview, Fusion Standard can route to on-device models such as MAI Code 1.1 Flash on Windows, or a manually configured local model in Copilot CLI.

Sources

1. Route and orchestrate models with GitHub Copilot Fusion (GitHub Changelog, 28 Oct 2026)
2. GitHub Copilot Fusion now supports selection of local models (GitHub Changelog, 28 Oct 2026)
3. GitHub Copilot Fusion orchestrates Anthropic and OpenAI models to balance performance and cost (Artificial Analysis, 28 Oct 2026)