Ramp Opens Its In-House LLM Router to Everyone, Citing 40% Cost Savings

Spend management company Ramp opened Router (Router.com), the model-routing infrastructure behind its own production AI traffic, to everyone on August 19, 2026¹. It provides one API to access models, evaluate them, and route traffic across them, built on three years of running the system on production workloads¹. The company says customers have cut costs by 40% without compromising quality or uptime¹.

Four routing strategies are available: Flex tier, which sends requests to a model’s discounted tier only when its observed latency matches the standard tier; shadow models, which send a sample of real requests to a candidate model while the current model keeps serving users; benchmark routing, which ranks models by up to three benchmarks and their relative weights; and NVIDIA Switchyard, which keeps routine agent turns on a lower-cost model and escalates the harder ones¹. Routing is free through the end of 2026, with model usage billed separately¹. The 40% figure is Ramp’s own accounting, and the announcement gives no denominator or comparison baseline. The layer that abstracts model choice drew other attention the same week, with reports that Stripe had reached a deal to acquire OpenRouter.

Sources

1. Introducing Ramp Router: Change AI Models Without Rebuilding - Ramp blog (August 19, 2026)
2. Ramp launches its own AI model router, called Router - TechCrunch (August 20, 2026)