Ramp has launched Router, an AI model routing service that says it can put multiple frontier models behind one endpoint and cut inference costs by 40 percent on average. The company is pitching one key, one bill, and a routing layer tuned for cost, performance, and fallback behavior, with free routing through 2026 and a $26 credit at launch. TechCrunch says the product is available in the United States for now and that users still pay underlying model inference costs. TechCrunch Router

The timing has a mild electrical crackle. Hacker News picked up Router by Ramp on August 19, one day before TechCrunch’s write-up, and the comment thread noticed the obvious overlap with the current scramble around model gateways. Ramp’s own materials frame Router as an internal tool that spent three years in production before becoming a product, with a claim of 2.75 trillion plus tokens routed monthly and a 30 percent cost reduction on internal workloads. HN thread Router

There is, however, a little static in the copy. Router’s homepage says "every model" more than once, while the currently listed set is large but not literally every model. Its FAQ also says Router stores model inputs, outputs, and metadata by default, while offering controls around retention and bring your own key for select providers. The page is admirably direct about all of this, which is more than many launch pages manage. Router FAQ

For Ramp, this looks less like a side quest than a familiar instinct translated into tokens. Spend management has met inference management, and apparently it wanted router.com while it was at it.