GITHUB'S CODER NOW PICKS ITS OWN MODEL — AT HALF THE FRONTIER PRICE

GitHub made Copilot Fusion generally available on October 28 at its Universe conference in San Francisco. It is not a new model but a runtime orchestration mode inside Copilot: developers pick an effort level — Standard, High or Max — and Fusion decides which model handles each coding task, using a single model, a cascade that escalates a cheap draft to a stronger one, or a critique in which one model reviews another's work.

Independent tester Artificial Analysis ran Fusion Max through Copilot CLI on its Coding Agent Index and scored it 60, two points behind Claude Code with Opus 5.5 Max at 62, at an average $5.59 per task against $11.42 — roughly half the cost and half the tokens, 7.2 million per task. Fusion led Terminal-Bench 4.0 at 71 but trailed on FrontierCode, so the near-parity is a composite, not a win on every kind of coding job.

The bigger savings number is GitHub's own: up to 77% on its DeepSWE-based tests, not the independent average. Users are billed for each underlying model Fusion calls at that model's rate, with a temporary 30% discount for paid subscribers — the pitch is completed work for less money, not a cheaper model.