On 28 October 2026, at its GitHub Universe conference, GitHub made Copilot Fusion generally available. Fusion looks at each task and decides which model or combination of models should handle it, then returns a single answer. It works in Visual Studio Code, the Copilot CLI, the GitHub Copilot app and JetBrains IDEs.
Why GitHub built it
Coding agents have become normal on GitHub very quickly. GitHub's Octoverse 2026 report says developers started more than 40% of pull requests with an agent in September, up from 17% in March, and GitHub describes that figure as a floor because some agent use goes undetected. The same report says 61% of Copilot users who chose their own model used more than one in September. People are already mixing models by hand, so GitHub has automated the choice.
The three moves Fusion can make
GitHub's documentation describes three execution patterns, and Fusion picks one for each task:
- Single, where one model does the whole job.
- Cascade, where an efficient model drafts a solution and a judge decides whether to accept it or escalate to a stronger model.
- Critique, where one model drafts, several models critique the draft in parallel, and the original model revises it using their feedback.
You also choose an effort level. Fusion Standard balances cost, quality and speed. Fusion High adds more orchestration with moderate token savings, and Fusion Max aims for frontier quality while still watching token use. Which levels you can use depends on your Copilot plan, so check the plan table in GitHub's docs.
What GitHub claims, and what to keep in mind
In its Universe announcement, GitHub says Fusion Max scored within two points of Claude Opus 5.5 Max on the Artificial Analysis Coding Agent Index, at half the cost and with 40% less time per task. That is a strong claim, and it comes from GitHub itself, so it is worth testing against your own codebase before you change how your team works. GitHub's docs also note a few practical details. Fusion has no context window of its own, so it works within the limits of the smallest model it might use, and file edits from a discarded draft are not automatically undone.
How billing works
There is no separate charge for Fusion. You pay for each model it uses at that model's standard rate, which means High and Max can use more AI credits on a single task because they may call several models. Through 30 November 2026, users on paid Copilot plans get a 30% discount on model costs when they use Fusion. GitHub also announced that Enterprise customers can now use every Copilot feature without a separate subscription and pay only for the AI credits they use.
What to do this week
Run a small, fair test. Pick three real tasks from your backlog, such as a bug fix, a small feature and a refactor. Run each one on the model you normally use and again on Fusion Standard, then compare the credits used and how much you had to fix by hand afterwards. If Standard holds up, try High on the hardest task. Our view is that routing like this will become the normal way people use AI at work, because very few people want to memorise model names, but your own results should decide when you switch.
For a wider view of coding assistants, see Compare, and for the model Fusion was benchmarked against, read which Claude model to use.