Ramp is expanding beyond corporate expense management with Router, a new AI model routing service built to help users and companies move requests across multiple large language models through a single API.
The launch puts Ramp into a growing part of the AI infrastructure market: tools that sit between customers and model providers, helping route inference requests by cost, performance needs, availability and testing preferences.
What Router Does
Router lets customers use and switch between large language models through an API. Ramp says it has relied on the router it built for its own AI usage needs over the past three years, and is now turning that internal system into a product.
The service currently provides access to models from OpenAI, Anthropic, DeepSeek, Moonshot, Minimax, Nvidia, xAI, and Z.ai. That gives users a way to work with several AI model providers without building separate switching logic for each one.
In practical terms, Router is designed to become a control layer for AI inference. Instead of treating every model call as a fixed choice, companies can define how requests should move between providers and models depending on what they want to optimize.
The product is similar in function to OpenRouter, though the source notes that OpenRouter offers many more AI model options than Ramp currently does. Ramp is therefore entering a category that already has a recognizable point of comparison, while tying its offer to its broader focus on spend management.
Pricing And Availability
Router is available only in the United States. Ramp is making the service free to use for the remainder of 2026, but users still have to pay the underlying AI model inference costs.
The company is also offering a $26 credit at launch. Ramp did not say how much Router will cost next year, which leaves the longer-term pricing model unresolved for customers evaluating it now.
That distinction matters because Router itself and model inference are separate cost layers. Free access to the routing service can lower the barrier to testing, but teams still need to monitor what they spend when models process tokens.
Ramp already has products related to AI token usage monitoring and token spend management. Router fits naturally beside those tools because it gives customers another place to view, control and direct AI usage.
Routing Strategies And Controls
Router includes several strategies for deciding where AI requests should go. One option lets users set a preference for model providers' flex usage tiers. Another lets Router choose which model should receive a query based on up to three user-specified benchmarks.
Users can also choose to send only difficult problems to expensive models. That creates a way to reserve higher-cost model usage for cases where a customer believes it is justified, while sending simpler work elsewhere.
The service is also positioned as a testing environment. Users can test models without having to switch manually, which can make comparison easier when teams are evaluating model behavior, cost, latency or fallback patterns.
Those controls point to the main appeal of model routing: companies do not always want one model for every task. Some requests may need speed, others may need a specific provider, and some may justify more expensive processing. A router gives teams a centralized place to express those preferences.
- Access models from several providers through one API.
- Route requests using selected strategies.
- Prefer flex usage tiers where available.
- Choose models based on up to three benchmarks.
- Send difficult problems to expensive models.
- Test models without repeated switching work.
Spend Visibility And Data Retention
Router includes a dashboard for monitoring AI usage. Users can see token spend, cost, latency, fallback attempts and other details.
That kind of visibility is especially relevant for companies using multiple models. Once AI usage spreads across providers, it can become harder to understand where costs are coming from, how often fallbacks happen, and whether latency is changing across different routing choices.
The dashboard gives Router a direct connection to Ramp's existing expense management identity. For customers already using Ramp to track and manage spend, AI inference becomes another category that can be monitored and controlled.
Router also comes with an opt-out data retention policy. By default, it will record model inputs, outputs and tool calls for one year. Ramp says it will remove "personally identifiable information before using that content to improve the product."
For companies evaluating the service, that retention policy is likely to be an important review point. The default is not no retention; users must understand the policy and decide whether to opt out.
Why Ramp Wants This Market
For Ramp, Router creates two linked opportunities. The company can participate in the AI inference market, while also offering current customers a model routing product that connects to token usage monitoring and token spend management.
The service could also become a place where users compare and test models. If Router becomes attractive as a model testing arena, Ramp may be able to build longer-term relationships with AI labs and inference providers worldwide.
That could give Ramp another path to new customers and another entry point for selling its expense management products. The company raised $750 million at a $44 billion valuation in June, and Router gives it a product that sits closer to how companies are actually using AI models day to day.
The central bet is straightforward: as organizations use more AI providers, they need better ways to route requests, watch costs and test model choices. Ramp is now trying to make that routing layer part of its broader financial operations platform.