GitHub has made Copilot Fusion generally available, turning its coding assistant from a model picker into a system that picks the models itself. Announced at GitHub Universe on October 28, Fusion routes each coding task across models, so developers choose an effort level rather than a specific model.
There are three effort levels. Fusion Standard favors versatile models for everyday coding, Fusion High favors more capable models for harder tasks, and Fusion Max favors the most capable models for the most demanding work. Under the hood, Fusion can run a single model, escalate from a cheaper model to a stronger one when a quality check calls for it, or have one model draft code while another reviews it and the first revises.
GitHub’s own benchmark claim is striking: on the Artificial Analysis Coding Agent Index, Fusion Max scored 60, within two points of Claude Code with Opus 5.5 Max, while using about half the API cost and tokens and completing tasks roughly 40% faster. Those figures are company-reported and await independent verification, but they frame the pitch clearly: frontier-quality results without paying frontier prices on every task.
Billing follows the models actually used, each charged at its standard rate, with a limited-time 30% discount on model costs for paid Copilot plans using Fusion. GitHub also reported that agents opened more than 40% of pull requests in September, up from 17% in March, a sign of how quickly agent-written code is becoming normal.
Developers comparing assistants can explore CodeRabbit, Codeium, Claude Code, and Gemini Code Assist in our directory to see how each one handles review, completion, and agentic coding.
Sources: FourWeekMBA, Particle News
