Ramp Launches Router.com for Free AI Model Routing
Corporate-spend platform Ramp has launched Router.com, a free model gateway that automatically routes API requests to the cheapest capable AI model to slash developer inference costs.
On August 19, 2026, corporate-spend platform Ramp, which powers over $200 billion in annual purchases, launched Router.com. The service is a single API endpoint that directs AI requests to the lowest-cost model meeting a developer's quality bar. Routing is free through 2026, with users paying list price for tokens and new users receiving $26 in credits. Early customers have cut inference costs by 40% on average, with Valentin De Matos of Delphi reporting a 92% reduction. The tool is a public version of internal systems Ramp built three years ago, which cut its own costs by roughly 30% while maintaining 99.9%+ reliability. The platform currently routes over 2.75 trillion tokens monthly, reflecting a broader trend where business AI spending on Ramp has grown 20.7x since June 2025.
Developers connect via an OpenAI- and Anthropic-compatible API to access 27 models, including Claude Opus 5, GPT-5.6 Sol, GPT-5.4 Nano, DeepSeek V4 Flash, and Grok. The launch coincided with Grok 4.6 reaching general availability on Amazon Bedrock. The gateway reaches models from OpenAI, Anthropic, and SpaceXAI, with Gemini support coming soon. Open-weight models like Nvidia, Kimi, DeepSeek, GLM, and Qwen are served via Fireworks AI, with Google, AWS, Together AI, Baseten, and Crusoe on the way. The routing engine applies over 100 optimizations across caching, compression, timing, and request handling, with automatic fallback.
To guide routing, the system uses Ramp SWE-Bench, a benchmark built from real production tasks. The benchmark shows that Claude Opus 5 solves tasks at an average cost of $1.84 per run, while Qwen3.7 Plus reaches a comparable solve rate at $0.15, and GPT-5.4 Nano handles cheaper work at $0.09. For practitioners, this enables automated, real-time cost arbitrage. Katelyn Lesse of Anthropic noted Claude is available from day one. Switching requires only a one-line base-URL change for OpenAI or Anthropic SDK users, with no Ramp card required. The service is initially limited to U.S. developers and offers zero-data-retention options on U.S.-hosted infrastructure.
This is our own summary of reporting by Unite.AI
