Cursor Router Uses a Trained Classifier to Cut Model Costs by 30-60%

Cursor introduced 'Cursor Router', a request-level classifier trained on over 600,000 real user requests to automatically select the optimal model per task. It routes simple tasks to cost-effective models and complex tasks to frontier reasoning models, reducing costs by 30-60% while maintaining quality. The system is cache-aware, accounting for prompt cache invalidation costs when switching models. In Intelligence mode, user satisfaction is near Fable 5's level with 60% lower cost. Cursor uses online metrics like user satisfaction and code keep rate for evaluation, abandoning offline evals. The feature is available for Teams/Enterprise with admin controls over model selection tiers.