Cloudflare has released Auto Router in public beta for AI Gateway, letting applications send requests to one model name while the gateway chooses the underlying model. The router aims to reserve expensive frontier systems for difficult work and use cheaper options when they are capable enough.
A classifier running on Cloudflare’s edge assigns each request to one of 14 categories and rates complexity, ambiguity, stakes and dependence on earlier context. A scoring matrix combines those signals with model benchmarks and token prices. It also accounts for cache costs in long conversations, where switching models can force the full context to be written again.
On an internal benchmark covering 97 simulated workplace tasks with three trials each, Auto Router completed 252 of 291 runs, or 86.6 percent, for $2.10. Claude Opus 5.5 completed 281 for $5.91, while GPT-6 Sol completed 245 for $2.64. Cloudflare says internal use through its OpenCode setup produced savings of up to 30 percent compared with relying only on frontier models.
Those results show a quality-cost trade-off rather than universal superiority: Opus completed more tasks. The beta is free, and Cloudflare plans to add zero-data-retention filtering, provider-capacity signals, reasoning-level selection and more models. Organizations should validate routing against their own workload and risk tolerance.