Cursor Launches AI Router to Cut Coding Costs by Up to 50%

AI coding platforms face a persistent trade-off between model capability and inference expense, particularly for companies processing large volumes of developer requests. Cursor is addressing that tension with Cursor Router, a system that evaluates each prompt’s complexity and domain before directing it to the most suitable AI model. The approach is designed to reserve expensive frontier models for demanding work while using lower-cost options where they can deliver comparable results.

Cursor has now formally released the request-level classifier for team and enterprise customers. The company said the router was trained on more than 600,000 real-world requests and can preserve frontier-level coding quality while reducing AI usage costs by 30% to 50%. At the upper end of that range, businesses could halve model spending without manually selecting a model for every development task.