GitHub Copilot Fusion leaves research preview for general availability
GitHub says its multi-model routing feature for Copilot scores within two points of Claude Opus 5.5 Max on a coding benchmark, at about half the cost.
GitHub shipped Copilot Fusion today, taking the multi-model orchestration feature that ran as a research preview under the name HydraFusion out of preview and into general availability. Instead of picking a model, developers choose an effort level, Fusion Standard, High or Max, and Fusion decides which models handle the task and how they work together.
Fusion routes work through one of three patterns: a single model solving the task directly, a Cascade that escalates from a cheaper model to a stronger one only if a quality check fails, or a Critique where a second model reviews the first’s draft before it revises. On the Artificial Analysis Coding Agent Index, GitHub says Fusion Max scored 60, within two points of Claude Code running Opus 5.5 Max, while using about half the API cost and token count and taking roughly 40% less time per task. It’s available on every Copilot plan in VS Code, the Copilot app, Copilot CLI and JetBrains IDEs, with a temporary 30% discount on model costs for paid plans.
Why it matters: picking a model for a coding task has mostly been a manual guess, and the usual failure mode is forgetting to switch back from an expensive model once the hard part is done. If GitHub’s own benchmark holds up outside its test harness, letting a router choose per prompt gets close to frontier quality without defaulting to the priciest model for everything.
The caveat: the benchmark comparison is GitHub’s own, specific to one benchmark revision and model pool, and billing still passes through each underlying model’s standard rate once the discount period ends.