Github Launches Copilot Fusion and New Copilot App, Cutting API Costs in Half

WHO:

The New Stack - GitHub announced Copilot Fusion, a system for selecting and orchestrating models for coding tasks.

The New Stack - Hari Pokhrel, principal software engineer at AT&T, said Fusion improved token efficiency and reduced unnecessary spending while preserving capacity for higher-value work.

The New Stack - GitHub also announced the GitHub Copilot app, Copilot Projects, and Copilot Automations.

The New Stack - The first article’s headline identifies Meredith Shubel as the author and The New Stack as the publication.

WHAT:

The New Stack - GitHub Copilot Fusion became generally available; it combines runtime model orchestration from Project HydraFusion with Copilot Auto Model Selection.

The New Stack - GitHub reported Copilot 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 - GitHub said Fusion Max achieved comparable benchmark performance at about half the average API cost per task, used about half as many tokens, and took about 40% less time per task than Claude Code with Opus 5.5 Max.

The New Stack - Developers choose among Fusion Standard for everyday coding, Fusion High for challenging tasks and deeper reasoning, and Fusion Max for the most demanding work.

The New Stack - GitHub said enterprise administrators can restrict which models Fusion selects, independently adjust each model’s policy state, and decide whether Fusion is the default for Copilot users.

WHERE:

The New Stack - GitHub announced Copilot Fusion at GitHub Universe, its flagship developer event in San Francisco.

WHEN:

The New Stack - GitHub said Fusion had improved since Project HydraFusion was released as a research preview in September 2026.

WHY:

The New Stack - GitHub’s stated vision is to make GitHub “the most comprehensive software delivery system in the world” by letting developers delegate model choices while retaining responsibility for the work’s direction and outcome.

The New Stack - GitHub said routing model selection to Fusion can help developers balance performance and cost and avoid choosing an unnecessarily powerful model for routine tasks.

HOW:

The New Stack - Fusion analyzes a task, selects a model, and can orchestrate work through a single model, a cascading workflow that checks and can escalate a response, or a critique workflow in which one model reviews another’s work.

The New Stack - Pokhrel said developers had often chosen a more powerful model for a complex task and forgotten to switch back for everyday activities.

The New Stack - GitHub said Fusion can choose different models for different workflow steps; developers may instead use enterprise controls to limit its choices.

The New Stack - GitHub Universe’s other announced products included the GitHub Copilot app, Copilot Projects, and Copilot Automations.

The New Stack - The article says Ramp moved model selection into its Open SWE open-source coding agent, while Replit made its intelligent model-routing system the default across accounts.

(UPDATED October 29, 2026 04:09 UTC)

GitHub has made Copilot Fusion generally available, a system that selects models for coding tasks and can route work through single-model, escalation, or review workflows. GitHub says its Fusion Max option delivered comparable benchmark performance to Claude Code with Opus 5.5 Max while using about half as many tokens, costing about half as much per task, and finishing about 40% faster; developers can choose Standard, High, or Max, while enterprise administrators can limit model choices. Announced at GitHub Universe in San Francisco alongside the Copilot app, Copilot Projects, and Copilot Automations, Fusion aims to balance performance and cost by choosing models suited to each task.