GitHub says agents started more than 40% of pull requests in September, up from 17% in March, and makes Copilot Fusion generally available

GitHub said more than 40% of pull requests on its platform started with an agent’s first draft by September, up from 17% in March, according to its 2026 Octoverse report published Wednesday (Oct. 28). GitHub said that amounted to 110 million pull requests in September alone, about six times March’s volume. It defines an agent-started pull request as one whose first draft came from an agent it can detect, and says its agent signals are a floor because not all agent use may be detected. GitHub said agent-started pull requests changed multiple files about 87% of the time, against 79% for pull requests without detected agent involvement, and that 90% of agent pull requests merged within two weeks in September, against about 86% for the comparison group.

The report said developers pushed code 17.9 billion times in the 12 months to Sept. 30, more than in 2024 and 2025 combined, with 34.3 billion commits and 1.3 billion merged pull requests. GitHub said 55% of monthly active Copilot users now use agentic features, up from 35% a year ago, and that 61% of Copilot users who chose their own model used more than one in September.

At its Universe conference the same day, GitHub made Copilot Fusion generally available. GitHub describes it as a system that chooses and coordinates models for each task rather than a single model, with Standard, High and Max effort levels. It is available on all Copilot plans in Visual Studio Code, the Copilot app, the command line interface and JetBrains IDEs. GitHub said users are billed at each underlying model’s standard rate, with a 30% discount on model costs for a limited time on paid plans.

GitHub said Fusion Max scored 60 on the Artificial Analysis Coding Agent Index, within two points of Claude Code with Opus 5.5 Max, at about half the average API cost per task and about 40% less time. Artificial Analysis, which tested Max through the Copilot command line interface on version 2.0 of its index, reported $5.59 per task against $11.42 for Claude Code with Opus 5.5 Max, and 7.2 million tokens per task, about half. It said Fusion led Terminal-Bench 4.0 at 71 but ranked third on FrontierCode at 51, behind 56 for Claude Code and 54 for Devin Fusion, and that its 57 on SWE-Atlas-QnA-AA lagged. VentureBeat noted the costs are benchmark API costs, not an estimate of what an enterprise will pay, and that Artificial Analysis’s article does not report the timing data behind GitHub’s 40% figure.

GitHub also said Enterprise customers can use all Copilot features without a separate subscription, paying for the AI credits they use, with self-serve access to follow. The company’s Universe post and The Next Web describe the announcements. Microsoft, GitHub’s parent, added a GitHub Copilot-powered coding tool to its own Copilot app in September, and Artificial Analysis separately ranked Gemini 4 Carbon first on its Intelligence Index on Wednesday.

GitHub says agents started more than 40% of pull requests in September, up from 17% in March, and makes Copilot Fusion generally available

Sources