Ramp launches Router, an LLM model router

Ramp, the finance-automation company, launched Router, a standalone service that lets developers and companies call and switch between large language models through one API. Through Router you can reach models spanning OpenAI, Anthropic, xAI and DeepSeek, plus Moonshot, MiniMax, Nvidia and Z.ai, with routing “strategies” that pick a model per request by cost, latency or quality. Ramp says it built the router for its own AI use over the past three years before packaging it as a product. It is free to use for the rest of 2026, with a $26 launch credit, and is US-only for now; customers still pay the underlying inference costs. Ramp says early users cut inference spend by 40% on average.

What it means for you

A model router is the practical answer to a problem builders now hit constantly: the best model for a task changes week to week, and rewiring your integration every time a cheaper or stronger model ships is friction you do not need. Router joins a crowded field, most visibly OpenRouter, which Stripe just acquired for a reported $7B, so this is a market with real money behind it. One caveat worth reading before you route production traffic: Router defaults to recording model inputs, outputs and tool calls for a year to improve the product, with an opt-out and a stated PII-removal step. That is the opposite posture from providers now competing on retention, like OpenAI’s zero-data-retention offering for frontier models. If your prompts carry sensitive data, set the retention policy before you send the first request.