Ramp launched Router, an API-based AI model routing service that lets users and companies access and switch among large language models through a single endpoint. Customers using Router have reduced inference costs by 40% on average, while Ramp says it cut its own costs by approximately 30% for equivalent output over three years of internal use. The service is available only in the United States, is free through the remainder of 2026 and includes a $26 launch credit, although users pay the underlying model inference costs; Ramp has not disclosed pricing for 2027. Router currently offers models from OpenAI, Anthropic, DeepSeek, Moonshot, Minimax, Nvidia, xAI and Z.ai. Its routing strategies can prioritize provider usage tiers, select models against user-defined benchmarks, send difficult tasks to more expensive models or simplify model testing. A dashboard tracks token usage, costs, latency and fallback attempts. Router also has an opt-out data-retention policy, recording inputs, outputs and tool calls for one year by default while removing personally identifiable information before using the data to improve the product. The service extends Ramp's existing AI token monitoring and spending controls and gives the company a potential foothold in the growing inference market. Similar to OpenRouter but with a narrower current model lineup, Router differentiates itself through its integration with Ramp's expense-management platform.