GitHub has launched Copilot Fusion, a system that handles model selection and orchestration for coding tasks, at its GitHub Universe event in San Francisco. The company says the feature lets developers delegate decisions about which models to use and how they work together, while keeping responsibility for the direction of their work.
GitHub debuted Copilot Fusion on Wednesday at GitHub Universe in San Francisco, the company's flagship developer event. The system combines runtime model orchestration from Project HydraFusion with Copilot Auto Model Selection, allowing it to both choose and coordinate models for reasoning, code generation, debugging and tool use.
The company positions the feature as a time-saver, saying developers no longer need to weigh performance and price for every task. Copilot Fusion can switch models mid-workflow based on what suits each step, using a single model throughout, a "cascading" workflow where a quality check evaluates a response and decides whether to accept or escalate it, or a "critique" workflow where one model reviews another's work.
GitHub reports that Fusion Max scored 60 on the Artificial Analysis Coding Agent Index, within two points of Claude Code with Opus 5.5 Max, while delivering that result at about half the average API cost per task, using about half as many tokens and taking about 40% less time per task.
Developers choose an effort level for each task: Fusion Standard for everyday coding, Fusion High for more challenging tasks and deeper reasoning, and Fusion Max for the most demanding work. Enterprise administrators can control which models Copilot Fusion may choose, disabling specific models for the feature without removing them from the whole team, and can set whether Copilot Fusion is the default for their users.
In a statement supplied to The New Stack, Hari Pokhrel, principal software engineer at AT&T, said: "Previously, developers would often choose a more powerful model for a complex task and forget to switch back for everyday activities." Pokhrel credited Copilot Fusion Standard with improving token efficiency at AT&T and reducing unnecessary spending.
Copilot Fusion arrives alongside other GitHub announcements including the Copilot app, Copilot Projects and Copilot Automations. Similar model-routing features have appeared elsewhere: Ramp moved model selection into its Open SWE coding agent this month, and Replit made its "intelligent model routing" system the default across every account in the summer.
1/ We announced our biggest update to Copilot to date, bringing together Chat, Cowork, Autopilot, and Code into one unified experience:
· Daily users grew nearly 50% QoQ, and more than 5X YoY.
· User satisfaction improved nearly 20% QoQ to another all-time high
· Number of customers with more than 50K Copilot seats, up 8X YoY
Great momentum with GitHub Copilot too, including Copilot Fusion, announced just this morning, which delivers frontier level quality, at up to 77% lower cost.
And this is just the beginning. The next frontier is maximizing intelligence per task while continuously optimizing for cost and performance.
This is the frontier that will matter to companies!
And we are extending our vision of Copilot as the new OS for work, bringing the business logic and capabilities of D365 directly into Copilot, helping every organization stand up their own infinite SaaS factory.
Developers can offload model choice and orchestration to GitHub's system, reducing the mental load of picking an efficient model for each task.
The claimed cost and time savings could change how teams budget for AI-assisted development.
Copilot Fusion reflects a broader industry shift toward automated model routing, following moves by Replit and Ramp.
Background
GitHub Copilot, launched as an AI pair programmer, has expanded into a suite of coding tools. Project HydraFusion began as a research preview in September 2026, and Copilot Fusion generalizes that work into a system that picks and orchestrates models automatically. The announcement was part of GitHub's stated vision to become "the most comprehensive software delivery system in the world," with model-level delegation while developers retain responsibility for the outcome of their work.
Key Perspectives
GitHub: says delegating model selection and orchestration to Copilot Fusion lets developers balance performance and cost without friction, freeing time and attention for other work.
Hari Pokhrel, principal software engineer at AT&T: says developers previously chose powerful models for complex tasks and forgot to switch back for everyday activities, and credits Fusion Standard with improving token efficiency and reducing unnecessary spending at AT&T.
What to Watch
Whether Copilot Fusion's claimed cost and token savings hold up in real enterprise workloads beyond benchmark tests.
How administrators respond to the ability to restrict which models the system may choose.
Whether competing coding agents follow Replit and Ramp in making model routing the default experience.
12 claims in this story checked against 1 source: 10 supported.
Show each claim
SupportedGitHub launched Copilot Fusion to automate model selection for coding tasks.The New Stack: “GitHub has debuted GitHub Copilot Fusion, a new system that handles model selection and orchestration for coding tasks so developers don’t have to.”
Partly supportedCopilot Fusion picks and switches models mid-task, promising lower cost and faster results than manually choosing models.The New Stack: “GitHub says Fusion Max delivered comparable benchmark performance to Claude Code with Opus 5.5 Max at about half the average API cost and in about 40% less time per task.”
SupportedHari Pokhrel said, “Previously, developers would often choose a more powerful model for a complex task and forget to switch back for everyday activities.”The New Stack: ““Previously, developers would often choose a more powerful model for a complex task and forget to switch back for everyday activities.””
SupportedGitHub debuted Copilot Fusion on Wednesday at GitHub Universe in San Francisco.The New Stack: “GitHub announced the news on Wednesday at GitHub Universe in San Francisco, the company’s flagship developer event.”
SupportedGitHub says developers can delegate model choices and coordination while retaining responsibility for the work’s direction.The New Stack: “with developers delegating model choices while retaining responsibility for the direction and outcome of the work.”
SupportedCopilot Fusion combines Project HydraFusion’s runtime orchestration with Copilot Auto Model Selection for reasoning, code generation, debugging and tool use.The New Stack: “Combining the advanced runtime model orchestration from Project HydraFusion with Copilot Auto Model Selection, Copilot Fusion can now handle both model selection and orchestration.”
SupportedCopilot Fusion can use one model, a cascading workflow, or a critique workflow, and can switch models mid-workflow.The New Stack: “That may be a single model doing everything, a “cascading” workflow where a quality check evaluates one model’s response and then decides to accept it or escalate it, or a “critique” workflow where one model reviews another’s work.”
SupportedGitHub reports that Fusion Max scored 60 on the Artificial Analysis Coding Agent Index, within two points of Claude Code with Opus 5.5 Max.The New Stack: “On the Artificial Analysis Coding Agent Index, GitHub reports that Copilot Fusion Max scored 60, within two points of Claude Code with Opus 5.5 Max.”
SupportedGitHub says Fusion Max delivered comparable benchmark performance at about half the average API cost per task.The New Stack: “GitHub says Fusion Max delivered comparable benchmark performance to Claude Code with Opus 5.5 Max at about half the average API cost and in about 40% less time per task.”
SupportedGitHub says Fusion Max used about half as many tokens.The New Stack: “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% less time per task.”
SupportedGitHub says Fusion Max took about 40% less time per task.The New Stack: “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% less time per task.”
Partly supportedRamp moved model selection into its Open SWE coding agent this month.The New Stack: “This month, Ramp described moving model selection into Open SWE, its open-source coding agent.”
