Superpower Daily
Netflix’s GenRec Pairs a 0.006% Gain With Catalog-Bound LLM Ran...Judge Vacates Pentagon’s Anthropic Blacklist and Orders Defense...Databricks Files $5B Equity Offering With Nearly All of It Alre...Snowflake Open-Sources Semi-Persistence With Sub-Second Single-...Cisco Rolls Out MyAgent to 90,000 Employees for Supervised Cros...Midjourney Opens V8.2 Image-Editing Tests With Up to Four Refer...Netflix’s GenRec Pairs a 0.006% Gain With Catalog-Bound LLM Ran...Judge Vacates Pentagon’s Anthropic Blacklist and Orders Defense...Databricks Files $5B Equity Offering With Nearly All of It Alre...Snowflake Open-Sources Semi-Persistence With Sub-Second Single-...Cisco Rolls Out MyAgent to 90,000 Employees for Supervised Cros...Midjourney Opens V8.2 Image-Editing Tests With Up to Four Refer...
Productspublished

Ramp Opens Its AI Router to U.S. Customers, With a One-Year Data Default

The new service gives companies one place to direct requests across eight model providers, but its U.S.-only rollout, future pricing and default retention policy set clear boundaries around the offer.

By 3 min read
Ramp Opens Its AI Router to U.S. Customers, With a One-Year Data Default
Sign in to save and react

Listen to this story

The audio brief

About 1:36
0:001:36
Read transcript
Ramp is opening its internal AI model router to U.S. customers, giving companies one API for sending requests across eight providers. The service can connect to OpenAI, Anthropic, DeepSeek, Moonshot, Minimax, Nvidia, xAI, and Z.ai, so customers do not have to switch model integrations by hand. Ramp says it has used the underlying system internally for three years. The key product is the routing decision. Companies can favor a provider’s flex-usage tier, choose up to three benchmarks, or reserve more expensive models for harder requests. They can also test different models without manually changing connections. A dashboard tracks token spend, total cost, response latency, and fallback attempts—showing not just which model handled a request, but the tradeoff between price, speed, and reliability. The launch terms are unusually clear, and limiting. Router is available only in the United States, and its catalog is smaller than OpenRouter’s. Ramp is waiving its service fee through the end of 2026, although customers still pay inference costs, and they receive a twenty-six-dollar launch credit. What the service will cost in 2027 has not been disclosed. The biggest governance issue is data retention. Inputs, outputs, and tool calls are kept for one year by default. Ramp says it removes personal information before using that content to improve the product, but customers must opt out if they do not want the default retention. So the immediate question is whether the controls and eventual pricing make Router practical beyond its limited first rollout.

Story brief

3 key points

Ramp is turning an internal model-selection system into a customer-facing routing layer, with U.S. access and free service through 2026 but no disclosed price afterward. Router can choose among providers using benchmarks, usage tiers, and task difficulty, while exposing spend, latency, and fallback data. The sharpest tradeoff is governance: prompts, outputs, and tool calls are retained for a year by default, with...

  1. 01

    Service fees are waived through 2026, with customers still paying inference costs and receiving a $26 launch credit.

  2. 02

    Router currently connects to OpenAI, Anthropic, DeepSeek, Moonshot, Minimax, Nvidia, xAI, and Z.ai.

  3. 03

    Customers can select up to three benchmarks, favor flex-usage tiers, or reserve pricier models for harder requests.

Ramp has launched Router, a U.S.-only service that lets companies send requests to and switch among multiple large language models through one API. The product puts an expense-management company closer to the operational decision behind AI spending: which model receives a request, at what cost, and with what performance tradeoff.

Router currently provides access to models from OpenAI, Anthropic, DeepSeek, Moonshot, Minimax, Nvidia, xAI and Z.ai. Rather than requiring a customer to separately change model connections, the service is designed to route requests across those options through its API.

Ramp says it used the underlying router internally for its own AI needs for three years before releasing it. The public launch extends a capability that fits alongside the company’s existing AI token-usage monitoring and token-spend management products.

Routing is the product

The service offers several ways to make that routing decision. A customer can prefer a provider’s flex-usage tier, select up to three benchmarks for Router to use in choosing a model, or reserve more expensive models for harder problems. It also supports testing models without manually switching between them.

What customers can monitor

  • Token spend and total cost, giving the routing choice a direct budget view.
  • Latency, the time associated with a model response.
  • Fallback attempts, showing when the system needed another path for a request.

That dashboard makes Router more than a model catalog. It combines the choice of model with measurements that can reveal the price, response-time and fallback behavior associated with that choice.

The important control is outside the router

The most consequential product setting may be its data policy. Router retains model inputs, outputs and tool calls for a year by default. Ramp says it removes personally identifiable information before using that content to improve the product, but customers that do not want the default retention must opt out.

Router’s pricing is also temporary. Ramp is offering a $26 launch credit and waiving its service fee through the end of 2026, excluding inference charges. The company did not state what Router will cost in 2027, leaving the longer-term economics unresolved for customers evaluating it as a routing layer.

Ramp enters a market where OpenRouter already performs a similar function, though OpenRouter offers more model options than Router currently does. Ramp’s narrower initial catalog, U.S. limitation and undisclosed post-2026 price make the launch a bounded first offering rather than a complete answer for every model-routing use case.

Sources

  1. techcrunch.comRamp launches its own AI model router, called Router | TechCrunch

Related coverage

Continue reading