Cloudflare releases AI Gateway Auto Router in beta to optimize AI spend

Cloudflare has released Auto Router in public beta via AI Gateway, designed to automatically route requests to the most cost-effective model capable of handling the task. By setting the model to cloudflare/auto, organizations can reduce token spend without requiring users to manually select models. In internal tests using a general knowledge work benchmark, cloudflare/auto achieved an 86.6% success rate at a total cost of $2.10 and a cost per success of $0.0084. The router uses a two-stage process: a classifier on Workers AI analyzes request complexity and category, and a scoring matrix ranks available models based on benchmark results, real-time capacity, and cache status. The system optimizes for cost-per-success rather than just price-per-token, leveraging the "jagged frontier" of model capabilities. The beta is free, with future plans to expand the model pool and add support for the Responses API and WebSockets.