Cursor Router Cuts AI Coding Costs 30–60% via Request-Level Model Selection

Cursor launched Cursor Router, a trained classifier that selects the optimal model per request, reducing costs 30–60% while preserving frontier quality. Trained on 600K+ real requests with user satisfaction as reward, it routes based on query, context, complexity, and domain. It is cache-aware, accounting for prompt cache misses. Three modes: Intelligence (near-Fable satisfaction, 60% cheaper), Balance (higher satisfaction than Opus 4.8, 36% cheaper), and Cost. Evaluation uses online metrics—user satisfaction and code keep rate—instead of offline benchmarks. Available for Teams/Enterprise with admin controls.