Route and orchestrate models with GitHub Copilot Fusion
By Steven Van ·
Fusion routes each prompt across models at three effort levels on all Copilot plans, with a 30% model-cost discount on paid plans through November 30.
GitHub Copilot Fusion is now generally available. In the model picker, GitHub users can choose Fusion instead of a specific model, and Copilot decides which models to use for each prompt and how they work together. Fusion combines Project HydraFusion's multi-model orchestration with Copilot auto model selection, as described in the announcement.
GitHub reports that in the Artificial Analysis Coding Agent Index, Fusion Max scored 60, within two points of Claude Code with Opus 5.5 Max. It did so at about half the average API cost per task, using about half as many tokens and taking about 40% less time per task.
Effort levels
After selecting Copilot Fusion, the user picks an effort level:
- Fusion Standard: balances cost, quality and latency, and favours versatile models for everyday work.
- Fusion High: favours capable models for challenging, well-scoped tasks such as a complex bug or a change across several files. It can make additional model calls, which may take longer and use more AI credits.
- Fusion Max: favours the most capable models for the most demanding tasks and complex reasoning, while still considering token efficiency.
A fourth level, Fusion Fast, is used only where fast models are needed. It cannot be selected by users and is available only in Visual Studio Code inline chat.
Execution patterns
For each prompt, Fusion picks one of three patterns, so different prompts in the same session can use different ones.
- Single: one model completes the task.
- Cascade: an efficient model drafts a solution, and a judge accepts it or escalates to a stronger model.
- Critique: one model drafts a response, multiple models critique it in parallel, and the original model revises it using the combined feedback.
Fusion also considers real-time system health and availability when choosing models. Users can follow the progress of each step and see which model is running. Only the final response is shown, since intermediate drafts may be revised or discarded. Hovering over the effort level name below a response lists the models used.
Availability and billing
Fusion is generally available in Visual Studio Code, Copilot CLI, the GitHub Copilot app and JetBrains IDEs, and works with all Copilot plans. It chooses only from models available under the user's plan and policies. If none of those models are available, Fusion does not appear in the model picker.
There is no separate charge for Fusion. Usage is billed per model at each model's standard rate, so Fusion High and Fusion Max can consume more AI credits than Fusion Standard. Through November 30, 2026, users on paid Copilot plans get a 30% discount on model costs when using Fusion.
Administrator controls
Administrators control which models Fusion can use and whether it is the default. Fusion only uses models enabled in enterprise and organization model policies. Enterprise administrators can view the models Fusion picks from and disable a model for Fusion without disabling it for everyone.
Limitations and local models
Fusion does not start subagents. If it discards a draft, file edits that draft already made in the workspace are not undone automatically, so changes should be reviewed before committing. Each step runs within the context limit of its model, and the context window shown for Fusion is a conservative value based on the smallest limit among the models it uses.
In public preview, Fusion can also use locally installed models at the Fusion Standard level: MAI Code 1.1 Flash (Windows) and, in Copilot CLI, Qwen3.8 Flash Next (Windows, macOS and Linux). This requires turning on experimental features.
