GitHub Copilot Fusion is a model-routing layer for agentic coding that replaces the manual model picker inside Copilot. Shipping in late October 2026, it is built for developers who want the right model chosen per task instead of setting one model globally. It sits beneath the standard Copilot experience and complements single-model tools like Cursor and Claude Code.
Core Features
- Three modes: pin one model, run a cheap-first cascade, or use a draft-and-review loop.
- Routes simple completions to small models and reserves large models for complex reasoning.
- Works across the GitHub Copilot surface: IDE, CLI, and the coding agent.
- Inherits Copilot's enterprise controls, audit logs, and repository context.
- Tracks which model produced each change, keeping an audit trail for compliance teams.
- Honors existing policy controls, so routing cannot bypass an org's allowed-model list.
Use Cases / Best For
- Teams watching token spend who want cheaper models on routine edits.
- Reviewers who prefer a draft from one model and a critique from another.
- Platform leads standardizing routing policy across a whole org.
Pricing
Copilot Fusion is included in all GitHub Copilot plans, including Copilot Individual at $10/month and Copilot Business at $19/user/month. See our AI programming tools list.
Our Take
Best for cost-conscious teams already on Copilot who dislike model micromanagement. The trade-off is that routing adds opacity: when a bug slips through, tracing which model produced it takes an extra step.
