2 hours ago
22
Ramp built its name helping companies track corporate spending, but the fintech just made a move that puts it squarely inside one of the hottest fights in tech right now: who gets to sit between businesses and the AI models they rely on. On Wednesday evening, Ramp introduced Router, a new Ramp AI model router that lets companies switch between different large language models through a single API, positioning the company as a fresh contender in the fast-growing AI inference market.
Key takeaways
- Ramp launched Router on August 20, 2026, an AI model routing service that lets users switch between large language models through one API.
- Ramp says it quietly used the same routing technology internally for three years before releasing it publicly.
- Router supports models from OpenAI, Anthropic, DeepSeek, Moonshot, Minimax, Nvidia, xAI, and Z.ai.
- The service is free through the rest of 2026, with a $26 launch credit, though users still pay for model inference costs.
- Router is currently limited to the United States and keeps model inputs and outputs on file for a year under an opt-out data policy.
Ramp launches Router to take on AI inference costs
Router answers a problem that’s become common for any company juggling multiple AI vendors: switching between models usually means rewriting code, managing separate contracts, and losing visibility into what’s actually being spent. Ramp’s pitch is that Router removes that friction by giving businesses one API through which they can tap into a whole menu of language models instead of locking into a single provider.
Three years of internal testing before public debut
This isn’t a rushed side project. Ramp says it built the routing system for its own internal AI usage and has been running it for three years before opening it up to outside customers. That kind of runway suggests the company treated Router less as a marketing stunt and more as infrastructure it trusted enough to depend on for its own operations before selling it to others.
What Router actually offers users
Router works much like OpenRouter, the AI routing platform that Stripe agreed to acquire just a day earlier, though Ramp’s version currently supports a narrower set of models. Still, the feature set is built for companies that want control over cost, speed, and quality all at once.
Model lineup spans OpenAI, Anthropic, DeepSeek and more
Through Router, customers can access models from OpenAI, Anthropic, DeepSeek, Moonshot, Minimax, Nvidia, xAI, and Z.ai. That’s a meaningful spread across US, European, and Asian AI labs, giving companies a way to compare and mix providers rather than betting everything on one lab’s roadmap or pricing.
Routing strategies and a cost-tracking dashboard
Router doesn’t just pass requests along blindly. It includes several routing “strategies” that let companies set rules for how queries get distributed. One option lets users prioritize a provider’s flex usage tiers to save money. Another allows Router to automatically pick a model based on up to three benchmarks the customer specifies. Businesses can also send only their hardest, most complex problems to pricier top-tier models while routing simpler tasks elsewhere, or test different models side by side without rewriting any code.
To back all of this up, Ramp built a dashboard that shows token spend, per-query cost, latency, and fallback attempts, giving finance and engineering teams a shared view of what their AI usage is actually costing them in real time.
Availability, pricing and data retention
Router is only available in the United States for now, which limits its immediate reach even as demand for AI cost-management tools grows globally. Ramp hasn’t said when, or if, it plans to expand the service internationally.
Free through 2026, but only in the US
On pricing, Ramp is leading with a straightforward hook: Router is free to use for the remainder of 2026, and new users get a $26 credit as a launch incentive. That doesn’t mean AI usage is free, though — customers still have to pay for the underlying model inference costs charged by whichever provider they route to. Ramp hasn’t disclosed what it plans to charge for Router once 2026 ends, leaving open how the company intends to monetize the product long term.
One detail worth flagging for enterprise buyers: Router uses an opt-out data retention policy. By default, it records model inputs, outputs, and tool calls for a full year, though Ramp says it strips out personally identifiable information before using any of that data to improve the product. Companies that want to avoid the year-long retention window will need to actively opt out.
Why Ramp is betting on model routing
Ramp’s decision to build Router isn’t happening in a vacuum. It comes just one day after Stripe confirmed it would acquire OpenRouter, the company widely seen as the category leader in AI model routing, in a deal reportedly worth $7.5 billion according to sources who spoke to the New York Times. That timing puts a spotlight on how valuable model-routing infrastructure has suddenly become to companies that already sit close to enterprise money flows.
For Ramp, the strategic logic is fairly direct. The company already sells AI token usage monitoring and token spend management tools to its corporate clients, so a routing layer slots naturally into that existing product stack rather than requiring it to build a new customer base from scratch. It’s a way to capture value from the booming AI inference market while deepening the relationship it already has with finance teams managing AI budgets.
There’s a second, less obvious upside too. If Router becomes a place where companies regularly test and compare AI models — the way OpenRouter did before Stripe’s acquisition — Ramp could end up building direct relationships with AI labs and inference providers around the world. Those relationships could become a new channel for selling Ramp’s core expense management products, turning a routing tool into a lead-generation engine for the rest of the business.
The move also lands at a moment when Ramp has plenty of capital to defend its position. The company raised $750 million at a $44 billion valuation in June, giving it the financial runway to compete for attention in a space where Stripe just showed it’s willing to pay a premium to own the routing layer outright.
FAQ
What is Ramp’s Router service?
Router is an AI model routing service launched by Ramp that allows users to switch between various large language models through a single API.
Which AI models can be accessed through Router?
Access to OpenAI models is provided by Router, alongside offerings from Anthropic, DeepSeek, Moonshot, Minimax, Nvidia, xAI, and Z.ai.
Where is Router available to users currently?
Router is currently only available to users in the United States.
How does Ramp price Router service initially?
Router is free to use for the remainder of 2026 with a $26 credit; however, users must still pay for AI model inference costs.
Article produced with the assistance of artificial intelligence and reviewed by the editorial team.
English (US) ·