GitHub Copilot Fusion is now generally available, ending the need for developers to pick a single AI model for each coding task. Announced at GitHub Universe on October 28, 2026, Fusion lets Copilot intelligently route and orchestrate models for each prompt, choosing the right model, or combination of models, while the developer simply chooses how much effort the task deserves.
The Model Picker Problem
Choosing a model used to be a manual trade-off. A strong model might be best for a hard bug, but wasteful for a routine edit, and developers often picked a powerful model for a complex task and forgot to switch back for everyday work. Fusion removes that decision from the developer’s day. When you select Fusion in the model picker, you are not selecting a specific model at all; you are selecting a system that routes and orchestrates models behind the scenes.
Fusion brings together two earlier pieces of work: Project HydraFusion’s multi-model orchestration and Copilot’s automatic model selection, with its fine-tuned model choice. The combination is meant to balance performance and cost efficiency, rather than optimising for one at the expense of the other. For each prompt, Fusion analyses what the task needs across reasoning, code generation, debugging and tool use, then decides how to execute it.
Developers control the effort, not the routing. Fusion Standard prioritises versatile models for everyday coding. Fusion High prioritises capable models for challenging tasks and deeper reasoning. Fusion Max prioritises the most capable models for the most demanding tasks and complex reasoning. That effort dial replaces a long list of model names with a single, understandable choice.
Three Ways To Run A Task
Inside Fusion, three workflows decide how models work together. In the Single workflow, one model solves the task directly. In Cascade, an efficient model drafts a response, and a quality check decides whether to accept it or escalate to a stronger model. In Critique, one model drafts an initial response, another model reviews it, and the original model revises its work.
Early evidence comes from AT&T, which GitHub quotes as a Fusion Standard user. Hari Pokhrel, a principal software engineer at AT&T, said Fusion Standard improved token efficiency by automatically selecting the right model for each task, reducing unnecessary spend and preserving capacity for higher-value work. He added that its decision-making has improved over time, giving developers more confidence to let Copilot handle model selection.
The Benchmark Claim, In Context
GitHub’s headline benchmark comes from the Artificial Analysis Coding Agent Index. On that index, Fusion Max scored 60, within two points of Claude Code with Opus 5.5 Max. GitHub says Fusion Max delivered that result at about half the average API cost per task, using about half as many tokens and taking about 40 percent less time per task.
Those figures are a single benchmark, not a guarantee for every repository, and teams should still test Fusion on their own codebases before assuming the same savings.
Billing follows the models Fusion actually uses. Users are billed for each model it invokes, at each model’s standard rate, and users on paid Copilot plans receive a limited-time 30 percent discount on model costs when using Fusion. That discount softens the cost of trying the new system, but it is temporary, and the long-run economics will depend on each team’s mix of Standard, High and Max work.
Controls, Clients And Rollout
Administrators keep control over which models Fusion can use. Fusion only draws on models enabled in an enterprise or organisation’s model policies, and administrators can view what models Fusion picks from, set each model’s effective policy state independently, and disable models for Fusion without disabling them for everyone. They can also set Fusion, at any effort level, as the default model for users under Enterprise managed settings.
Fusion is available on all Copilot plans in Visual Studio Code, the GitHub Copilot app, the GitHub Copilot CLI and JetBrains IDEs, with support for additional clients coming soon. Users who previously enabled auto, HydraFusion or HydraFusion Max during the research preview will transition automatically to the corresponding Fusion effort levels.
Why It Matters
Fusion matters because model selection has become a hidden tax on coding agents. As more models arrive with different strengths, prices and speeds, asking every developer to become a model router wastes attention and money. A system that routes well can make expensive AI-assisted development economical, and that economics question now sits at the centre of the coding-agent market, from Cursor’s point-and-fix design tools to Bloomberg’s AI coder inside BQuant finance.
The risk is the mirror image of the benefit. Routing decisions are only as good as the evaluation behind them, and a system that quietly picks a cheaper model for a task that needed a stronger one can cost more in rework than it saves in tokens. Fusion’s Cascade and Critique workflows are meant to catch that failure, but developers should still review the result, not just the price. If Fusion earns trust on real codebases, the model picker itself may fade away; if it does not, developers will keep reaching for a specific model they trust.
Background
GitHub Copilot began as an in-editor code completion tool and has grown into an agentic coding platform spanning editors, the command line and a standalone app. Fusion, previewed earlier as HydraFusion, is its move from single-model assistance to orchestrated, multi-model coding work.
Source: GitHub changelog
0 Comments