The provider routing system used by OpenRouter now allocates traffic based on a balance of inference reliability and cost, rather than price alone. The update to the routing algorithm is designed to award more requests to AI inference providers that deliver higher quality output relative to their cost, which the company says will encourage a competitive environment for provider performance.
The company reported that early testing for open-weight models showed throughput and time to first token both increased by about 20%. Additionally, the transition to the new routing logic resulted in approximately 40% fewer fallbacks and about 60% fewer rate-limit errors.
