Ramp launches Router, an AI model-routing service

Three years of internal use have become a product. Ramp has launched Router, an API that lets companies send requests to and switch between large language models from multiple providers. The service is currently available only in the United States.

Router connects to models from OpenAI, Anthropic, DeepSeek, Moonshot, Minimax, Nvidia, xAI and Z.ai. Its routing strategies can favor providers’ flex usage tiers, choose a model using up to three user-defined benchmarks, or direct only difficult problems to more expensive models. Users can also test models without changing their integration manually.

The service adds a control panel to the model layer: customers can see token spend, cost, latency and fallback attempts. It is free for the remainder of 2026, and comes with a $26 launch credit, but customers still pay the costs of AI model inference. Ramp has not disclosed the price for next year.

The concrete change is for teams already managing AI usage through Ramp: model choice and model spending can sit alongside the company’s existing tools for monitoring tokens and controlling token costs. For Ramp, Router could also help build relationships with AI labs and inference providers, extending beyond its corporate expense platform.

The service is not without boundaries. OpenRouter currently offers many more AI model options, and Router’s opt-out data retention policy records inputs, outputs and tool calls for one year by default. Ramp says it removes personally identifiable information before using that content to improve the product. The launch gives customers a working routing service, but its broader reach and next-year pricing remain open.