Copilot Fusion lets you stop choosing a model in GitHub Copilot: you select an effort level—Standard, High, or Max—and Copilot decides which models to use for each task. It’s generally available as of October 28, 2026, across all Copilot plans, in VS Code, the GitHub Copilot app, Copilot CLI, and JetBrains IDEs.
Note: The pricing data in this article, including a discount ending November 30, 2026, reflect GitHub’s documentation at the time of publication.
What is Copilot Fusion?
Copilot Fusion is a real-time model router and orchestrator that replaces where a model used to sit in the model picker. When you select it, you’re not choosing a model: you’re choosing a system that decides, for each request, three things: what execution pattern to use, how much reasoning effort to invest, and which models to run. It also takes into account the system’s state and availability at that moment.
GitHub presents it as the union of two existing pieces: the multi-model orchestration from the research preview Project HydraFusion and Copilot’s automatic model selection (the auto mode).
What is GitHub Copilot and where does Fusion fit in?
GitHub Copilot is GitHub’s AI programming assistant: it completes code, answers in chat, and runs tasks as an agent within your editor, terminal, or its own app. Until now, each conversation ran on a model you chose in the picker. Fusion is another option in that same picker, but instead of a fixed model it gives you a system that chooses for you.
How does Fusion decide which models to use?
For each prompt it chooses one of three workflows:
- Single: a single model solves the task directly.
- Cascade: an efficient model drafts a solution and a judge accepts it or escalates it to a more powerful model.
- Critique: a model drafts, multiple models critique the draft in parallel, and the original model revises it with all the observations.
The pattern is chosen prompt by prompt, so a single session can combine all three. You only see the final response. Cascade and Critique take longer because they make additional calls to models.
Standard, High, or Max? Which to choose in Copilot Fusion
| Effort Level | What it’s for | How it behaves |
|---|---|---|
| Standard | Everyday programming | Balances cost, quality, and latency; chooses one model per task |
| High | Large but well-scoped tasks | Balanced orchestration with moderate token savings |
| Max | Deepest reasoning and analysis | Frontier capability, without losing sight of token efficiency |
A rule of thumb that emerges from the documentation: start at Standard, move to High when the task is large but clearly defined, and save Max for problems where a wrong answer costs more than the additional credits.
There’s also a fourth level, Fusion Fast, which you can’t select: it only works in VS Code’s inline chat.
How do you activate Copilot Fusion?
Update your client, open the model picker, choose Copilot Fusion, and select an effort level. That’s the entire procedure GitHub documents: there’s no separate flag or configuration key for individual users.
It works on:
- Visual Studio Code
- The GitHub Copilot app
- GitHub Copilot CLI
- JetBrains IDEs
GitHub says more clients will arrive, without naming them or giving dates. If none of the models Fusion uses are available to you, Fusion simply won’t appear in the picker.
Is Copilot Fusion free? How much it costs in GitHub Copilot
Fusion has no separate charge, but it’s not free either: GitHub charges you for each model Fusion uses, at that model’s standard rate. The feature is available across all Copilot plans.
That has a practical consequence: High and Max can call more models per task, so the same task can consume more AI credits than Standard. Fusion tries to keep the main conversation on a single model when it can, to take advantage of cached tokens.
Until November 30, 2026, Copilot’s paid plans get a 30 percent discount on model costs when using Fusion.
At the time of publication, GitHub’s documentation didn’t name the models that make up Fusion’s pool, didn’t explain how Fusion interacts with premium requests, and didn’t detail which effort levels each plan includes.
Is Fusion Max worth it?
GitHub reports that Fusion Max scored 60 points on the Artificial Analysis Coding Agent Index, two points behind Claude Code with Opus 5.5 Max, at roughly half the average API cost per task, around half the tokens, and about 40 percent less time per task. GitHub didn’t publish the methodology or test date alongside the figure, so take it as vendor data until you measure it in your own code.
What happened to HydraFusion?
HydraFusion is now Fusion. If you had auto, HydraFusion, or HydraFusion Max enabled during the research preview, your selection automatically moves to Fusion’s equivalent effort level; you don’t need to do anything.
If you followed our previous guide, How to use HydraFusion in GitHub Copilot CLI, the concepts still apply: what changed is the name and the entry in the picker.
Does Copilot Fusion work with local models?
Yes, in public preview and only at the Standard level. There are two compatible local models:
- MAI Code 1.1 Flash: Windows only; installed from Copilot CLI, the Copilot app, or VS Code.
- Qwen3.8 Flash Next: installed with a provider like Ollama; in Copilot CLI it appears as
qwen3.8-flash-next-local. Works on Windows, macOS, and Linux.
For Fusion to route to them:
- Copilot app: enable Local model auto routing in the Experimental section of settings.
- Copilot CLI: start it with
--experimental, or run/experimental onwithin an interactive session.
How does an administrator control Copilot Fusion?
Fusion only uses models that your enterprise and organization policies already allow. Enterprise admins can see which models Fusion chooses, define the policy for each model for Fusion separately, and disable a model just for Fusion without removing it from everyone. Data residency, FedRAMP, and models-in-evaluation policies also narrow Fusion’s pool.
To make Fusion the default model, the enterprise managed configuration (copilot/managed-settings.json in the enterprise’s .github-private repository) uses two keys: model with the value "fusion", and fusionTier, which sets the default effort level. Accepted values are "standard", "high", "max", and "unmanaged"; any value other than "unmanaged" locks the level against user or repository changes. This is GitHub’s example, which lets enterprise teams change the default value:
{
"model": "fusion",
"fusionTier": { "overridable": "standard" }
}
fusionTier requires Copilot CLI 1.0.104-3, VS Code 1.144.0, or the GitHub Copilot app 1.1.33-4, or later versions.
What should you keep in mind before using Copilot Fusion?
- A discarded draft can leave changes. If Fusion discards a draft that already edited files in your workspace, those edits don’t undo themselves. Review the diff before committing; if you want to isolate those changes, the GitHub Copilot local sandbox is a good additional layer.
- No sub-agents. Fusion doesn’t start sub-agents.
- The context window is the smallest in the pool. Fusion doesn’t have its own window: each step runs within the limits of the model that executes it, and the window shown is a conservative value based on the smallest limit among its models.
- It’s going to keep changing. GitHub warns that execution patterns and the model pool can change over time.
What about you? What specific task would you give Fusion Max first to see if it beats your usual model?
- Standard
- High
- Max
- I’d rather keep choosing the model myself (tell us which)
Comment below or, if this article reached you by email, reply directly to the email: your reply gets published here.