Ramp launches Router, its AI model routing API for cost-efficient inference

Ramp, the corporate expense management platform, has entered the AI inference arena with Router, a new model routing service that lets users and companies switch between large language models through a single API. The service, which Ramp says it has been using internally for three years, currently supports models from OpenAI, Anthropic, DeepSeek, Moonshot, Minimax, Nvidia, xAI, and Z.ai. Router offers several routing "strategies," including the ability to prioritize providers' flex usage tiers, route queries based on up to three user-specified benchmarks, or send only difficult problems to expensive models while using cheaper ones for simpler tasks. Users get a dashboard showing token spend, cost, latency, fallback attempts, and other metrics. The service is free to use for the remainder of 2026, includes a $26 credit at launch, and is available only in the United States for now.

For Ramp, Router represents a two-pronged opportunity. The company touched off after Stripe's similar move into AI inference, and it aims to tap the booming market while offering existing clients a natural extension of its AI token usage monitoring and spend management products. Router has an opt-out data retention policy: by default, it records inputs, outputs, and tool calls for one year, though Ramp says it removes personally identifiable information before using that content to improve the product. If Router gains traction as a testing ground like OpenRouter, Ramp could build long-standing relationships with AI labs and inference providers worldwide. That would help the company, which raised $750 million at a $44 billion valuation in June, acquire new customers and create a fresh entry point for its expense management offerings.