GitHub unveils Copilot Fusion and usage-based Enterprise access
GitHub’s Octoverse report said more than 40% of pull requests in September began with an agent’s draft, compared with 17% in March.

GitHub launched a system called Copilot Fusion at its Universe developer conference to assign AI models to different parts of coding tasks. The company also opened all Copilot features to GitHub Enterprise customers through AI-credit payments, without requiring a separate subscription.
Fusion assesses requirements for reasoning, generating code, debugging and using tools before selecting one model or several. Jay Parikh, Microsoft’s executive vice president of CoreAI, said Fusion Max came within two points of Anthropic’s Claude Opus 5.5 Max on the Artificial Analysis Coding Agent Index, at half the cost and using 40% less time per task.
Enterprise customers must initially arrange the usage-based option through GitHub’s sales team; self-service access is planned. Customers with Copilot Business or Copilot Enterprise subscriptions can stay on those plans or switch.
A separate demonstration featured the GitHub Copilot app, where agents can work simultaneously across multiple repositories. A cloud agent session can move among the app, VS Code, a command-line interface and the web. GitHub also said Copilot automations can learn from results and suggest changes for developer approval.
In its annual Octoverse report, GitHub said an agent’s draft was the starting point for more than 40% of pull requests in September, versus 17% in March. The company cautioned that it cannot identify every agent. The report also recorded 17.9 billion code pushes during the 12 months ending 30 September, more than the combined figure for 2024 and 2025.




