Ramp launches its own AI model router, called Router
Corporate expense management leader Ramp is making a strategic play into the AI infrastructure market, mirroring moves by fintech giants like Stripe to capitalize on the surging demand for AI inference. The company officially unveiled Router this Wednesday, a sophisticated API-based service designed to help organizations seamlessly navigate and switch between various large language models (LLMs).
A Proven Internal Tool Goes Public
Ramp is not entering this space as a novice; the company notes that it has relied on this proprietary routing technology to manage its own internal AI workloads for the past three years. By opening the service to its broader customer base, Ramp is positioning itself as a central hub for businesses looking to optimize their AI spending and performance.
Currently, the service is exclusive to United States-based users. To encourage adoption, Ramp is offering the service for free through the end of 2026, supplemented by a $26 launch credit. While users remain responsible for the underlying costs of AI model inference, the company has yet to disclose its pricing structure for the post-2026 period.
Key Features and Model Support
Router functions similarly to existing platforms like OpenRouter, providing a unified interface for interacting with a diverse array of industry-leading models. Current supported providers include:
- OpenAI and Anthropic
- DeepSeek, Moonshot, and Minimax
- Nvidia, xAI, and Z.ai
Beyond simple access, Router offers advanced "strategies" that allow companies to fine-tune their AI operations. Users can route queries based on specific benchmarks, prioritize cost-efficient "flex" tiers, or automatically direct complex tasks to more powerful, expensive models while routing simpler queries to cheaper alternatives.
"Router provides a dashboard that lets them see token spend, cost, latency, fallback attempts, and other details."
Data Policy and Strategic Vision
Transparency is a core component of the new dashboard, which tracks critical metrics such as latency and token expenditure. Regarding data privacy, Router operates under an opt-out data retention policy. By default, the system logs inputs, outputs, and tool calls for one year, though Ramp emphasizes that it strips personally identifiable information (PII) before utilizing any data to refine its product offerings.
For Ramp—which secured a $44 billion valuation following a $750 million funding round in June—this launch represents a dual-purpose growth strategy. It allows the company to capture value from the booming AI inference sector while deepening its integration with existing clients who already rely on Ramp for token usage monitoring and expense management. By establishing itself as a vital arena for model testing, Ramp is effectively creating a new entry point to expand its footprint in the enterprise software market.