GitHub used its Universe developer conference in San Francisco on Wednesday to launch Copilot Fusion, a router that decides which model — or which mix of models — handles each step of a coding task. It weighs what the step needs across reasoning, code generation, debugging and tool use, and can pick a single model, combine several, or skip a model call entirely when none is required.

The performance claim is unusually specific. According to Jay Parikh, Microsoft's executive vice president for CoreAI, "Fusion Max" lands within two points of Anthropic's Claude Opus 5.5 Max on the Artificial Analysis Coding Agent Index — at half the cost and with 40% less time per task. Satya Nadella put it on LinkedIn as frontier-level quality, "40% faster and at up to 77% lower cost".

None of that has been independently verified. What is verifiable is the strategic shift: GitHub is moving the competition from model quality to cost per completed task. A router that blends vendors turns price into the headline number, and a benchmark score into one input among several.

The second announcement is commercial. Under "Copilot Access", every Copilot feature is now included for GitHub Enterprise Cloud customers at $21 per user per month, with no separate Copilot licence; usage is billed as AI credits. Existing Copilot Business and Copilot Enterprise customers can keep their current plan or "move over when they are ready" — the migration is voluntary. Sign-up currently runs through sales, with self-service promised shortly. Copilot Standalone remains at $10 a month, but without an included credit allowance, so usage is billed from the first request.

GitHub also showed the Copilot app, where agents work across several repositories in parallel managed sessions, with continuity that carries a single cloud session between the app, VS Code, the command line and the web. Copilot automations can now learn from their own results and propose changes for developers to approve. A new Copilot code review Max is said to post the highest F1 score on ReviewBench for high-severity issues.

The Octoverse numbers explain why any of this is being rebuilt. In the twelve months to 30 September, developers pushed code 17.9 billion times — more than 2024 and 2025 combined. Commits reached 34.3 billion; merged pull requests 1.3 billion, triple the year before. More than 40% of pull requests in September began with a draft from an agent, up from 17% in March; that was 110 million in September alone. GitHub attributes the inflection to June, when agents began working across days and running in the cloud, and notes that its agent count is a floor because it cannot detect every agent.

The volume is the story. GitHub reported 1.4 billion commits a month in April and 2.9 billion in August. For September the figure is 7.4 billion, against roughly 65 million a month in 2024. The platform now counts more than 225 million developers, 38.3 million of them in the United States and 32 million in India. TypeScript remains the most-used language; Rust grew fastest among the top languages, doubling in a year.

That load has a cost. Outages have become a recurring theme at GitHub, and the company attributes them to agentic demand itself. It added 3 million CPU cores earlier this year and is adding 4 million more, plus 160 petabytes of high-speed storage, extra Actions compute and 50 Tb/s of network capacity. Underneath, the platform is being re-architected to separate compute from storage — internal benchmarks suggest up to 35x higher write throughput — and moved to a multi-region deployment on Azure, expected to complete early next year.

Parikh supplied the framing: "I think all the software we've written to date will amount to less than 1% of the software we'll write in the next five years." Whether that holds is unknowable. What is measurable is that a version-control service is being provisioned for a workload nobody seriously planned for two years ago — and that GitHub is rewriting its pricing at the same time.

Sources