GitHub has released Copilot Fusion generally available, presenting it as a “single option” that chooses which AI model—or combination of models—handles each part of a developer’s coding task. Announced at GitHub Universe, Fusion evaluates prompts to determine needs such as reasoning, code generation, debugging, or tool use, then routes work accordingly based on an “effort level” (Standard, High, or Max).

Sources also describe how Fusion can run in different modes: a direct single-model approach, a cascade where a cheaper model drafts and a check either accepts or escalates, and a critique setup where one model drafts, another reviews, and the first revises. GitHub and Microsoft frame Fusion’s value in cost and performance terms, including claims of lower per-task cost and faster results, alongside enterprise controls and model-cost billing details.

Other coverage emphasizes the broader context of agentic coding. The Next Web links the launch to the trend of AI agents contributing to software changes, noting its claim that agents start work on a substantial share of pull requests.