Cursor Launches Cursor Router – a Query Classifier to Cut AI Costs by Up to 50% Without Losing Quality

Cursor has made Cursor Router publicly available for Teams and Enterprise plans. It is a classifier that analyzes a request before invoking the model and routes it to the optimal model for the given task. According to the company, online A/B tests have achieved 60% cost savings while maintaining frontier-level quality, and three enterprise customers in early access saved 30–50%. The problem the system addresses is cost structure: about 60% of developers use a single model daily, spending on routine tasks at frontier-model prices. Cursor Router was trained on 600,000+ live requests and optimized for user satisfaction (AFC). For each request, it analyzes the query, context, task complexity, and domain. Simple tasks are routed to cost-effective models, UI updates to the model with the best taste, and complex long-term problems to frontier models. Savings come not from degrading complex task solutions but from moving routine work out of the frontier price category. An important detail: Cursor Router accounts for cache miss costs when switching models, which are included in the stated savings. The system supports updating as new models emerge. Two metrics are used for evaluation: user satisfaction and generated code retention rate. Auto Intelligence mode delivers satisfaction on par with Fable at 60% lower cost, while Auto Balance exceeds Opus 4.8 at 36% lower cost. Cost per commit: Auto Balance $4.63, Auto Intelligence $6.76, Opus 4.8 $7.34, Fable 5 $12.69. Cursor Router works on desktop, web, iOS, CLI, and via the Cursor SDK, and is enabled by default for Teams. Administrators can configure enable/disable settings, mode restrictions, and allowed model lists. Two limitations: Grok 4.5 is required as a cost-effective option (blocklist cannot exclude it), and Balance and Intelligence modes are billed at the selected model's rate, so costs vary.