Ramp Launches Router.com to Automate AI Model Selection and Cost Control

Ramp launched Router.com on August 19, 2026, as a single API endpoint that sends AI requests to the lowest-cost model that satisfies a developer's quality standards. [1] The company offers free routing through 2026, charging users the list price for tokens and providing $26 in credits to new users. [1] According to the announcement, customers already utilizing the service have reduced their average inference costs by 40%. [1]

The product makes Ramp's three-year-old internal tooling public, which the company claims previously cut its own inference costs by approximately 30% while maintaining 99.9%+ reliability. [1]

The model picker currently includes 27 models, ranging from Claude Opus 5 and GPT-5.6 Sol to budget options like DeepSeek V4 Flash and GPT-5.4 Nano. [1] Developers connect via an OpenAI- and Anthropic-compatible API to reach models from OpenAI, Anthropic, and SpaceXAI. [1]

The routing engine uses over 100 optimizations, including caching, compression, and automatic provider fallback. [1] All models on the platform are hosted on U.S.-based infrastructure. [1]

Ramp is commercializing its own internal infrastructure to capitalize on the rapid growth of enterprise AI expenditures, which the Ramp AI Index indicates grew 20.7x since June 2025. By unifying access to 27 models across multiple providers—and with Gemini support announced as coming soon—Router.com positions itself as a central cost-control layer for developers facing unpredictable inference bills. The concurrent availability of Grok 4.6 on Amazon Bedrock highlights a broader industry push toward widespread model distribution across multiple access points. What the sources don't address: How Router.com will price its routing layer after the free period concludes at the end of 2026.