Ramp launches its own model router to compete with OpenRouter on inference layer

Ramp, the corporate spend-management platform, is entering the large-language-model routing market with Router, an API service that lets customers switch between LLM providers without changing code. The company says the system has been running in production for three years and is now opening it to external use, initially only in the United States.

Router includes a dashboard that displays token usage, cost, latency, fallback attempts and other operational metrics in real time. Pricing for the launch is free until the end of 2026 (customers pay only the providers’ inference costs) plus an initial credit of $26 (≈95 shekel). Ramp has not disclosed what the service will cost in 2027. The data-retention policy is opt-out: by default input, output and tool calls are stored for one year, with the company committing to remove PII before using the data to improve the product.

Supported providers include OpenAI, Anthropic, DeepSeek, Moonshot, Minimax, Nvidia, xAI and Z.ai, a considerably smaller offering than OpenRouter. The advertised advantage lies in built-in routing strategies: flexible usage tiers of providers, routing based on up to three benchmarks defined by the customer, sending complex queries only to higher-cost models, and the ability to test alternative models without changing integration.

Ramp raised $750 million in June (≈2.7 billion shekel) at a valuation of $44 billion (≈160 billion shekel). The model router serves two purposes. First, it adds a revenue layer from the hot inference market. Second, the service integrates with Ramp’s existing products for monitoring token usage and managing AI spend, creating a natural sales path for enterprise customers already managing compute budgets through the platform. If Router becomes a popular testing ground like OpenRouter, Ramp could build direct relationships with AI labs and inference providers worldwide, opening a door to sell its expense-management solutions to new customers.