OpenRouter has introduced a new Auto router that selects models using signals from how people already choose models across the platform. Instead of relying only on a hand-built task classifier, the router is informed by large-scale usage patterns from millions of decisions.
The company says this market-style approach performs better than conventional task-based classifiers across a broad range of prompts. For developers, the change is meant to reduce the guesswork of deciding which model to call when cost, speed, and quality vary by task.
The update reflects a wider shift in AI infrastructure: model selection is becoming a product feature of its own. A router can make multi-model systems easier to operate, but it also adds another layer that developers need to monitor. Teams using it should still evaluate results on their own workloads, because aggregate user behavior may not match a specialized application’s risk tolerance or quality bar.