Last updated: October 29, 2026
GitHub Copilot Fusion became generally available on October 28, 2026. Instead of choosing a model, you choose an effort level (Standard, High or Max), and Copilot routes each prompt to one or more models. It works in VS Code, Copilot CLI, the Copilot app and JetBrains IDEs, and paid plans get 30% off model costs through November 30.
Copilot Fusion merges the multi-model orchestration from the Project HydraFusion research preview with Copilot’s auto model selection, according to the GitHub changelog.
Contents
- What’s new
- How Copilot Fusion works
- Effort levels
- How to turn it on
- Billing and admin controls
- Compared with HydraFusion and other agents
- Limitations
- Frequently asked questions
What’s new in GitHub Copilot Fusion
Copilot Fusion replaces three separate picker choices from the preview period with one option and an effort setting. GitHub says that in the Artificial Analysis Coding Agent Index, Copilot Fusion Max scored 60, 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 per task (changelog). These are GitHub’s figures, not independent tests.
| Item | HydraFusion preview | Copilot Fusion |
|---|---|---|
| Status | Research preview | Generally available (Oct 28, 2026) |
| Picker options | Auto, HydraFusion, HydraFusion Max | Copilot Fusion with Standard, High or Max effort |
| Clients | VS Code, Copilot app, Copilot CLI | VS Code, Copilot app, Copilot CLI, JetBrains IDEs |
| Local models | Not part of routing | MAI Code 1.1 Flash or Qwen3.8 Flash Next on Fusion Standard (public preview) |
| Admin controls | Preview features toggle | Per-model Fusion policy and Fusion as the default model |
| Pricing | Each model at its standard rate | Each model at its standard rate, 30% off through Nov 30, 2026 |
How Copilot Fusion works
For each prompt, Copilot Fusion uses capability signals for reasoning, code generation, debugging and tool use to pick the cheapest execution pattern that meets the quality bar for your effort level. The documentation describes three patterns:
- Single: one model completes the task. It behaves like a normal request to one model.
- Cascade: an efficient model drafts a solution, and a judge decides whether to accept it or escalate to a stronger model.
- Critique: one model drafts while several models critique it in parallel; the feedback is combined and the original model revises.
The pattern can change from prompt to prompt in the same session. Reasoning effort and model choice are set per step, and routing happens at natural cache boundaries so you do not pay extra cache costs without a quality gain. It has no context window of its own: each step runs within the limits of the model running it, and the window shown in the picker is a conservative value based on the smallest limit in the pool.
GitHub Copilot Fusion effort levels
| Effort level | Priority | Typical use |
|---|---|---|
| Fusion Standard | Balances cost, quality and latency | Everyday work; selects one model per task |
| Fusion High | Balanced orchestration with moderate token savings | Complex bug fixes or changes across several files |
| Fusion Max | Frontier capability while still considering token efficiency | Tasks that need the deepest reasoning and most thorough analysis |
| Fusion Fast | Fast models only | Not selectable; used only in VS Code inline chat |
A practical pattern is to keep Standard as your default and switch to High or Max only for the hard turn in a session, then drop back. Because cascade and critique add review calls, High and Max take longer and can use more AI credits.
How to turn on GitHub Copilot Fusion
- Update your client. Copilot Fusion needs a current version of VS Code, the Copilot app, Copilot CLI or a JetBrains IDE.
- Open the model picker and select Copilot Fusion. In the Copilot app, turn on the Copilot Fusion switch in the model dialog.
- Choose an effort level: Fusion Standard, Fusion High or Fusion Max.
- Check the models used by hovering over the effort level name below the finished response.
- Review file changes before you commit, because edits made by a discarded draft are not undone automatically.
In Copilot CLI, use fusion as the model name and set the tier with --fusion-tier (CLI model docs):
copilot --model fusion --fusion-tier standard -p "YOUR PROMPT"To let Fusion Standard use a local model, install MAI Code 1.1 Flash (Windows, on supported hardware) or Qwen3.8 Flash Next through a provider such as Ollama, identified as qwen3.8-flash-next-local. This is a public preview: turn on Local model auto routing under Experimental in the Copilot app, or run /experimental on in Copilot CLI. GitHub says local routing lets tasks finish without using GitHub AI credits.
Billing and admin controls
There is no separate charge for Copilot Fusion. You pay for each model it calls at that model’s standard rate, and through November 30, 2026, paid plans get a 30% discount on those model costs (billing section). Fusion keeps the main conversation on the same model when it can, so cached tokens keep working, and helper models such as a reviewer get only the context they need.
Fusion only uses models allowed by your plan and by enterprise and organization model policies, including data residency, FedRAMP and evaluation-model restrictions. Enterprise administrators can disable a model for Fusion without disabling it for everyone, and set Copilot Fusion at any effort level as the default model through Enterprise managed settings. If none of its models are available to you, Copilot Fusion does not appear in the picker.
How Copilot Fusion compares
The HydraFusion research post from September 4 reported that, in offline tests, its best configuration improved verified task quality on TerminalBench 2.1 by 4.9 percentage points at 67% lower estimated cost than Claude Opus 5. The October 14 update added HydraFusion Max, a model pool including Claude Opus 5.5, GPT-6 Astra, GPT-6.1 Sol and GPT-6 Luna, medium effort for drafts and high effort for review, and parallel critics.
The main difference from a single-vendor agent is that you stop choosing models. If you prefer one model end to end, compare our Claude Code power-user guide and our GPT-6.1 Terra guide. For models you can run yourself, see our Mistral Large 4 guide and DeepSeek V4.1 Flash guide, and for Google’s latest model, the Gemini 4 Argon guide.
Limitations to know
- Copilot Fusion works separately from subagents and does not start them.
- If it discards a draft, file edits that draft already made stay in your workspace.
- The execution patterns and model pool can change over time, so results may shift between weeks.
- Local model routing is a public preview and applies only to Fusion Standard.
Frequently asked questions
What is GitHub Copilot Fusion?
GitHub Copilot Fusion is a runtime model router and orchestrator in GitHub Copilot. You pick Copilot Fusion and an effort level in the model picker, and it chooses which models to use, how much reasoning to apply and whether to add a review step, then returns a single response. It became generally available on October 28, 2026.
What is the difference between Copilot Fusion and HydraFusion?
HydraFusion was the research preview of multi-model orchestration. Copilot Fusion combines HydraFusion’s orchestration with Copilot auto model selection. Research preview users who had auto, HydraFusion or HydraFusion Max selected are moved to the matching Copilot Fusion effort level automatically.
Which Copilot Fusion effort level should I use?
Use Fusion Standard for everyday work, Fusion High for substantial, well-scoped tasks such as a complex bug fix or a change across several files, and Fusion Max for tasks that need the deepest reasoning. High and Max can use more than one model per task, so they can consume more AI credits.
How much does Copilot Fusion cost?
There is no separate charge. You are billed for each model Copilot Fusion uses, at that model’s standard rate. Through November 30, 2026, users on paid Copilot plans get a 30% discount on model costs when they use Copilot Fusion.
How do I see which models Copilot Fusion used?
After the response arrives, hover over the effort level name shown below the response to see the list of models used. While it works, Copilot also shows each step and the model running it.
Sources
- GitHub Changelog: Route and orchestrate models with GitHub Copilot Fusion
- GitHub Docs: About GitHub Copilot Fusion
- GitHub Docs: Choosing a model in GitHub Copilot CLI
- GitHub Changelog: HydraFusion updates and new Max option
- GitHub Blog: Project HydraFusion, frontier quality via multi-model orchestration
Image: GitHub mark (Octicons), by GitHub, MIT license, via Wikimedia Commons. Placed on a white background and resized.
