Ramp Launches 'Router': A Service That Lets You Switch Between AI Models in One Place

The corporate expense software company is now selling access to AI models from OpenAI, Anthropic, DeepSeek and others through a single service, free through 2026, with a $26 credit to start.

AI2Day NewsdeskAsistido por IAPublicado Updated Editor: Lee Brown3 min read
A close-up of a modern office telephone handset resting on a desk, with soft warm light and abstract blue-green waveform visualisations floating gently above it
Ilustración creada con IA. No es una fotografía de los hechos descritos.
Share

Key points

  • Ramp launched Router on Wednesday, a service letting businesses access multiple AI models through one connection point instead of juggling separate accounts.
  • Router offers models from eight providers, including OpenAI, Anthropic, DeepSeek, Nvidia and Z.ai.
  • The service is free through the end of 2026, though users still pay the underlying AI model costs.
  • Router records your model inputs and outputs for one year by default unless you opt out.
  • Ramp raised $750 million at a $44 billion valuation in June 2025.

Ramp, best known for helping companies track employee spending, has moved into the AI infrastructure market. On Wednesday it launched Router, which lets businesses and individual developers reach a collection of AI models through a single connection rather than signing up with each provider separately.

We covered the market Ramp is entering on 16 August, when Stripe was reported to be buying OpenRouter for more than $7 billion, the closest existing rival to what Ramp just built.

What exactly does Router do?

Router is an API gateway, a technical front door that sits between your software and a collection of AI models, sending each request to whichever model you've told it to use. The eight providers available at launch are OpenAI, Anthropic, DeepSeek, Moonshot, Minimax, Nvidia, xAI and Z.ai.

Routing decisions work a few ways. You can set a preference for cheaper, lower-priority processing tiers. You can let Router pick automatically based on up to three performance benchmarks you specify. Or you can reserve the expensive, capable models for genuinely hard problems and let lighter ones handle the rest.

A dashboard shows token spend (a count of how much text you've sent and received, which is how AI companies charge for usage), response speed and fallback attempts, meaning the times one model failed and Router tried another. Ramp says it's been running its own internal AI work through a version of this system for three years.

Should users be concerned about privacy?

Read the small print before signing up. By default, Router logs your model inputs, outputs and tool calls for twelve months. Ramp says it strips personally identifiable information before using that data to improve the product, but the logging is on unless you turn it off. Anyone processing sensitive business data through Router needs to check those settings first.

The contrast with OpenAI's approach is sharp: our 19 August story on Private Safety Processing described a system that scans conversations for misuse then deletes everything rather than keeping it for a year.

What does this cost?

Router is free to use through 2026, and new users get a $26 credit. You still pay the AI providers' normal model costs on top. Ramp hasn't said what it will charge in 2027.

The service is currently available only in the United States. OpenRouter carries a wider range of models and has been around longer, so Ramp faces a real competitor from the start.

For Ramp, the play is clear: sell AI model access to the same finance teams already using its expense tools, and build relationships with AI labs that could open new business lines. Router's 2027 pricing will be the real test of whether it can pull companies away from established alternatives.

What to watch for: Check Router's data retention settings before routing any sensitive queries through the service. The default is to keep your inputs and outputs for a year.

© 2026 AI2Day