What GitHub announced
GitHub’s October 28 changelog says Copilot Fusion is generally available across Copilot plans in VS Code, the Copilot app, Copilot CLI, and JetBrains IDEs. Choosing Fusion means selecting a routing and orchestration system rather than one named model. We have not established the announcement’s first public hour. [1]
Fusion can use one model, escalate a draft to a stronger model, or have models critique a response. The announcement describes Standard, High, and Max effort options. Available models and options still depend on subscription and administrator policies; “all plans” does not mean every plan has identical choices. [1][2]
Each model used can contribute to the bill
GitHub bills each model Fusion uses at that model’s standard rate, with no separate Fusion fee. Its documentation specifies a 30% model-cost discount for paid Copilot plans through November 30, 2026. High and Max can use multiple models and consume more AI credits than Standard. A discount is not a promise of a lower total task cost. [1][2]
A discarded draft can leave edits behind
GitHub’s documentation warns that discarding an intermediate draft does not automatically undo file changes that draft already made. Review the workspace changes, even when the final answer looks polished. Some local-model routing is a separate public preview; this announcement is not a blanket on-device processing guarantee. [2]
Our take
For a small development team, less model switching is useful only if the result is worth its total cost. Try a bounded task on a separate branch, compare the credits consumed, and inspect the diff before accepting the work. Extra reviewers can help; their presence alone does not establish correctness.
We reviewed public sources. We have not run Fusion or compared its performance.