Ramp Launches Router, Its Own AI Model Routing Service

Ramp, a corporate expense management platform, announced the launch of Router, an AI model routing service, on Wednesday evening. The service allows companies to access multiple large language models through a single API and switch between them based on specified criteria. Ramp has been operating Router internally for three years before opening it to paying customers, suggesting the company has tested the technology extensively within its own operations.

Router’s launch positions Ramp alongside other infrastructure companies recognizing AI inference as a strategic business opportunity. Ramp’s June funding round brought in $750 million at a valuation of $44 billion, providing capital for new product development. Stripe is reportedly acquiring OpenRouter, the existing market leader in model routing, for over $7 billion—an acquisition that underscores the commercial importance investors assign to this emerging market segment.

Access to Multiple AI Vendors

Router provides connections to models from eight major AI companies: OpenAI, Anthropic, DeepSeek, Moonshot, Minimax, Nvidia, xAI, and Z.ai. This selection is narrower than OpenRouter’s current offerings, which provide access to dozens of models across both leading vendors and smaller or experimental model creators. Ramp’s more limited initial lineup suggests the company is launching with its most critical vendor partnerships, potentially expanding coverage as Router gains adoption.

The service is currently restricted to the United States. For the remainder of 2026, Ramp is charging no service fee to use Router—customers pay only for the actual inference costs from their chosen model providers. New users receive a promotional $26 credit toward those inference costs. The company did not disclose what Router will cost starting in January 2027, leaving the long-term pricing structure uncertain.

Dashboard Metrics and Visibility

Router includes a dashboard providing visibility into inference spending and performance. The dashboard displays token usage counts, costs incurred, latency measurements for model responses, fallback attempt logs, and other operational data relevant to managing AI workloads. This monitoring capability aligns with Ramp’s core product—expense management—and suggests the company views Router as an extension of its existing token spend management features rather than as a standalone service.

How Routing Strategies Work

Read also: Binance Opens Trading to AI Agents With User-Controlled Limits · Calendly Launches Note-Taker Focused on Post-Meeting Automation Work · Reach Capital Closes $265M Fund V, Bets on AI for Human Potential

Cost Optimization Routes

Router offers multiple routing strategies, each designed for different organizational priorities. One strategy lets users specify a preference for models from vendors offering flex-rate pricing tiers, automatically directing requests toward those providers to reduce costs for unpredictable or variable workloads. This approach is particularly relevant for companies evaluating model usage patterns or operating with variable AI workload demands.

Another strategy allows organizations to configure routing rules based on query characteristics. Users can, for example, instruct Router to send complex computational problems to expensive, high-capability models while directing routine or straightforward requests to cheaper alternatives. This cost-differentiated approach requires teams to understand their own workload distribution but can yield significant savings for organizations with diverse AI needs.

Benchmarks and Automatic Selection

A third routing strategy enables automatic model selection based on up to three user-specified performance benchmarks. Organizations can weight factors such as response quality, latency speed, or accuracy on specific tasks, then let Router choose which model to invoke based on how each vendor’s offerings perform against those criteria. This approach shifts the optimization burden from the developer to the platform, making it easier for teams to scale AI usage across multiple models without manual intervention.

Model Testing and Experimentation

Router also supports simple A/B testing and model evaluation workflows. Teams can assess how different models perform on the same inputs without reconfiguring API endpoints or redeploying applications. This feature is valuable for organizations piloting new models or comparing vendor offerings before making long-term commitments to specific providers.

Data Retention and Privacy

Router operates under an opt-out data retention policy, recording model inputs, outputs, and tool calls for one year by default. The company stated it removes “personally identifiable information before using that content to improve the product,” though Ramp did not explain the technical or operational processes ensuring this removal occurs. The opt-out structure—where retention is the default and removal requires action—shifts the burden onto users to manage their data retention preferences.

For organizations handling sensitive business data or customer information, the default one-year retention period may warrant investigation and explicit configuration to align with internal data governance policies. The scope of what Ramp considers “personally identifiable information” subject to removal could meaningfully impact which internal data the company retains and uses for product development.

Why Ramp Is Building This

Ramp’s entry into model routing creates two distinct business opportunities. First, the company gains direct exposure to the rapidly expanding market for AI inference services. As organizations scale their AI usage, they increasingly need tools to manage multiple vendor relationships, compare costs and performance, and optimize routing decisions—a market that did not exist two years ago.

Second, Router integrates with Ramp’s existing product suite. The company already offers AI token usage monitoring and token spend management capabilities; Router extends these by letting customers optimize which model receives each request. For existing Ramp customers already using the platform to track AI spending, Router appears as a natural upgrade—a way to not only monitor spending but actively control it through intelligent routing decisions.

The alignment with Ramp’s core business in expense management differentiates Router from OpenRouter, which functions as a standalone service available to anyone. Ramp’s approach prioritizes quick adoption within its installed base, where integration with existing tools and workflows creates switching costs and stickiness.

Market Competition and Timing

The reported Stripe acquisition of OpenRouter for $7 billion or more signals that major infrastructure companies view model routing as strategically important. Stripe’s move—adding a routing capability to its payments and financial infrastructure platform—mirrors Ramp’s strategy of building products that complement its core business.

OpenRouter’s broader model catalog and established market position give it an initial competitive advantage, particularly for customers seeking access to cutting-edge or experimental models. However, OpenRouter’s independence means it must serve all customers equally, while Ramp’s integration with its own expense management platform allows preferential positioning within that ecosystem.

The launch of Router during a period when AI inference spending is accelerating, combined with increasing competition for this market segment, suggests Ramp sees model routing as essential infrastructure for its long-term positioning. How quickly Router accumulates users during its free tier period in 2026 will likely inform both Ramp’s commitment to the product line and its pricing strategy for 2027.