Cursor launches Router, an automatic model-selection system for coding requests

Cursor has launched Router, a feature that automatically picks which AI model handles each coding request instead of leaving that choice to the developer, aiming to cut costs without giving up frontier-model quality.

What's new

Cursor announced Router on July 22, 2026: "Cursor Router lets teams automatically route every request to the most capable model for the task, delivering frontier intelligence at a lower cost." Under the hood, it's a purpose-built classifier rather than a simple rules engine: "Cursor Router is a classifier that routes users to the best model option based on their query," trained and validated against thousands of live coding requests before rolling out broadly.

On results, Cursor says the approach holds up at scale: "In online A/B tests across millions of requests, Cursor Router delivered frontier-quality performance at 60% savings." The idea is that many coding requests — simple edits, boilerplate, straightforward completions — don't need a frontier-tier model to get a good result, while harder requests get routed to more capable (and more expensive) models automatically. Teams aren't limited to pure automation, either: users can bias the router toward Intelligence, Balance, or Cost depending on the task. The feature is live now: "Cursor Router is available today for Teams and Enterprise plans across desktop, web, iOS, CLI, and our SDK," with admins able to enable it per team, allow or block specific models, set defaults, and disable specific optimization modes.

Context

Cursor, like most AI coding tools, has historically left model selection to the developer — pick GPT, Claude, Gemini, or another supported model for a session and stick with it. That model works fine for developers who already have strong opinions about which model handles which kind of task, but it leaves cost and quality on the table for teams that don't want to micromanage model choice on every request, especially at Enterprise scale where usage volume and per-token cost compound quickly.

Router is Cursor's attempt to remove that decision from the developer entirely, treating model selection as an optimization problem the product should solve rather than a preference the user should manage.

Why it matters

Automatic model routing is becoming a standard pattern across the AI tooling stack — from consumer chat products to now developer tools and, separately, generative media platforms — as the number of viable, differently-priced models in any given category keeps growing. For a coding tool specifically, the pitch is direct: enterprises running large teams on Cursor face real per-seat AI spend, and a router that claims frontier-quality output at roughly 60% lower cost is a meaningful lever on that bill without asking teams to change how they work. It also puts Cursor in more direct control of the model layer within its own product — steering usage toward whichever models it judges best for a given task, including presumably its own commercial arrangements — rather than simply being a thin interface over whatever model the developer happens to pick.