By Hamza Faisal
GitHub has made Copilot Fusion generally available, the centrepiece of its annual Universe conference, repositioning its popular AI assistant from a code-completion tool into a system that orchestrates AI agents across the whole software development lifecycle. Instead of asking developers to pick a model, Fusion picks the model — and the strategy — for each task.
The routing engine, previously previewed under the name HydraFusion, evaluates what a job needs across reasoning, code generation, debugging and tool use, then dispatches it across a pool of frontier models including Anthropic's Claude Opus 5.5 and Sonnet 5.5 and OpenAI's GPT-6 Astra and GPT-6 Sol. Developers choose an effort level — Standard, High or Max — and Fusion handles the rest.
Near-frontier results at half the cost
GitHub's headline claim is efficiency. On the Artificial Analysis Coding Agent Index, Fusion Max scored 60, within two points of Claude Code running Opus 5.5 Max at 62 — but at about half the average API cost per task, $5.59 against $11.42, using half as many tokens and finishing roughly 40 percent faster. On the DeepSWE benchmark, GitHub said Fusion High and Max improved task completion by up to 8.7 points while cutting estimated costs by up to 77 percent. Those figures are company-reported and unverified.
Fusion picks between three coordination strategies for each request. "Single" hands the whole task to one model; "Cascade" has one model plan and another execute; "Critique" lets one model propose an approach while a second reviews and challenges it before the work is completed. There is no orchestration fee, and GitHub is offering a 30 percent discount on Fusion model costs through November 30.
Workflows written in plain language
Alongside Fusion, GitHub made Agentic Workflows generally available. The system, built by GitHub Next, lets teams automate reasoning-based jobs — issue triage, CI failure analysis, documentation updates, test writing — by describing the task in natural-language Markdown files that compile into standard GitHub Actions YAML. Workflows run inside GitHub Actions, reusing existing runner groups and policy controls, and can be powered by Copilot or a developer's own coding agent.
Agents run with read-only permissions by default inside a sandboxed container behind an Agent Workflow Firewall, with a threat-detection job scanning proposed changes before they are applied. One early adopter, Accenture, reported a team lead saving 29 hours a week, velocity more than doubling, AI authoring 91 percent of commits, and production errors falling from about 11,000 a day to under 1,000.
Built for an agent-driven flood of code
The announcements land as AI agents reshape GitHub itself. The company said agents opened more than 40 percent of pull requests in September, and the platform recorded 7.38 billion commits that month — five times the previous month's total — a surge that strained its systems and contributed to earlier outages. In response GitHub is adding more than 4 million CPU cores and 160 petabytes of storage and rebuilding its Git architecture for a possible 35-fold throughput gain.
Copilot is now available to all GitHub Enterprise users without a separate per-seat purchase, with billing moving to token-based credits and administrators controlling which models Fusion may use. New additions include Copilot Projects, Copilot Connectors, AI Scans for pull-request vulnerabilities, and a package firewall for npm fetches. The message from Universe is blunt: developers are shifting from writing code to directing the agents that do.



