The State of Model Routing — NVIDIA, Cognition, OpenRouter
Industry leaders declare model routing essential as frontier model costs become prohibitive
“We firmly believe that we're in a multi-model world.”
A panel featuring NVIDIA, Cognition (makers of Devin AI engineer), and OpenRouter discussed the emerging discipline of model routing — dynamically selecting which model to use per task based on cost and capability tradeoffs. NVIDIA noted it released NeMo Triton models with full open datasets, weights, and recipes to support customization in this multi-model world. Cognition reported that enterprise customers are increasingly asking how to measure model ROI and determine which tasks justify expensive frontier models versus cheaper alternatives, signaling that cost-aware routing is moving from nice-to-have to a core production concern.