Cloudflare Releases Auto Router in Public Beta for AI Cost Savings

Cloudflare has released Auto Router in public beta through AI Gateway, aiming to reduce AI spend by automatically routing requests to capable models. Early results show a cost savings of up to 30% compared to using frontier models. Auto Router uses a multi-head classification model to determine the best model for each task, balancing quality and price. It filters out models that do not support the request format or execution mode, and accounts for credentials, billing configuration, access control policies, and spend limits. The router considers the conversation, task categories, and model benchmark results to estimate how well each model fits the request. It then selects the model with the highest utility, combining expected quality with each model's input and output token prices.