Ramp opens its AI router to everyone
Ramp Router, which cut Ramp's own LLM costs 30% across 100+ features and routes 2.75 trillion tokens a month, is now free for outside developers to use.
Ramp opened public access to Ramp Router this week, an OpenAI-compatible endpoint that automatically sends each request to the lowest-cost model that still clears a quality bar. Ramp built the router internally three years ago to keep more than 100 AI features across its finance platform on the right model for each task, and says it cut Ramp’s own LLM spend by 30% while making those features faster. The router already handles 2.75 trillion tokens a month across Ramp’s products. It supports frontier models from OpenAI and Google along with select open models, including Kimi, and is free at launch, priced at list token rates, with the first 500 outside signups getting $100 in credits. No Ramp account or company card is required to use it.
The notable part is who is shipping this: Ramp is a corporate-card and spend-management company, not an AI infrastructure vendor, and it built model routing as internal cost-control tooling before deciding it was good enough to give away. That mirrors what Cursor did with Cursor Router days ago and what several other product teams have shipped as routing becomes a default feature rather than a differentiator: once a company runs enough AI traffic internally, building a router pays for itself, and open-sourcing or productizing it afterward is close to free marketing.
For builders choosing between LLM gateways, Ramp Router is worth evaluating alongside existing options specifically because it has three years of production traffic behind its routing decisions rather than launch-day defaults, though as with any new gateway, benchmark it against your own workload before routing anything cost-sensitive through it.