Choosing the right AI model for every coding task has become a chore, especially now that Copilot users can draw on dozens of them. GitHub Copilot Fusion, which became generally available on October 28, takes that decision off your plate: you pick an effort level and Fusion decides which models to use, and how to combine them, for each prompt. Here is how to set it up, when to use each mode and how to keep costs under control.
What Copilot Fusion Does
Fusion combines Copilot's existing automatic model selection with technology from Project HydraFusion, GitHub's research into multi-model coordination, which ran as a preview in Copilot CLI from September and later in VS Code and the Copilot app. Rather than locking a task to one model, Fusion picks a workflow per prompt:
- Single: one model solves the task directly
- Cascade: an efficient model drafts a response, and a quality check either accepts it or escalates to a stronger model
- Critique: one model drafts, another reviews, and the original model revises
GitHub reports that in the Artificial Analysis Coding Agent Index, Fusion Max scored 60, within two points of Claude Code running Opus 5.5 Max, at about half the average API cost per task and around 40% less time. Treat that as a vendor-reported result; the other effort levels have not been independently scored.
How to Turn It On
The steps are the same in every supported client:
- Update VS Code, the GitHub Copilot app, Copilot CLI or your JetBrains IDE to a version that includes Fusion
- Open the model picker and select Copilot Fusion
- Choose an effort level
GitHub says Fusion is available on all Copilot plans, though the models it can choose from are limited to those your plan and your organisation's policies allow. If you used auto, HydraFusion or HydraFusion Max during the preview, you have been moved to the matching Fusion level automatically.
Choosing Standard, High or Max
The effort level is your main control over quality, speed and cost:
- Standard prioritises versatile models for everyday coding, such as small fixes, explanations and routine refactors
- High prioritises more capable models for challenging, well-scoped tasks that need deeper reasoning
- Max reaches for the most capable models for the hardest problems, such as complex multi-file changes or tricky debugging
A sensible default is to start at Standard and step up only when a task stalls or the output needs repeated correction. Cascade and Critique workflows add extra review calls, so higher levels can cost more time and more credits.
Keeping Costs Predictable
There is no separate orchestration fee. You are billed for each model Fusion calls, at that model's standard rate. For a limited time, paid plans get a 30% discount on model costs when using Fusion; check GitHub's announcement for the exact end date. Because one prompt can trigger several model calls, watch usage closely during your first week rather than assuming Fusion is always cheaper.
Tips for Teams and Admins
Enterprise administrators keep control. Fusion only uses models enabled in your model policies, and admins can view the models Fusion may pick from, disable a model for Fusion alone without removing it for everyone, and set Fusion at any effort level as the default through enterprise managed settings.
Before switching a whole organisation over, run a short pilot:
- Compare Fusion with your usual fixed-model setup on the same representative tasks, prompt and repository state
- Record pass or fail, effort level, models used, elapsed time and cost for each run
- Work in a branch or disposable worktree and keep agent permissions tightly scoped
- Review the full diff, including tests, and treat model-generated critique as advisory, not a replacement for human code review
If you need a fixed sequence of steps or human checkpoints, a defined workflow is a better fit than letting Fusion choose dynamically.
Why It Matters
GitHub's own Octoverse data shows why a router like Fusion has arrived now. In September, 61% of Copilot users who picked a model themselves used more than one, and paid plans drew on 47 different models, with none taking more than a quarter of requests. Developers are already mixing models by hand; Fusion automates that habit and adds structured review between models. The risk is that routing makes cost and behaviour less transparent, which is why measuring results per task matters more than ever.
The Bottom Line
Copilot Fusion turns model selection from a manual guessing game into a single effort setting, and its Cascade and Critique modes bring multi-model review to everyday coding. Start on Standard, escalate deliberately, measure cost per task, and let admins set guardrails before making it the default.
