Google adds model routing; Cloudflare unifies AI gateway controls

Google Cloud API Gateway now supports rules-based model routing in public preview. One OpenAI-compatible endpoint can route requests to Gemini, Claude, or OpenAI open-weight models hosted in Google's Model Garden, while application authentication stays separate from backend model credentials. 1

The preview is deliberately narrow: one router's backends must share a Google host, routing reads only the request's model field, OpenAPI 3.x is required, and VPC Service Controls are not supported. 2

Cloudflare's immediate release is a unified control plane for Workers AI and external providers. A default gateway can be created on the first authenticated request, adding logs, token counts, latency, errors, and cost attribution; AI Gateway credits can now cover Workers AI usage. Provider failover and prompt-based smart routing remain roadmap items, not current capabilities. 3

Try it today with a sandbox: put two text models behind one gateway endpoint, send the same prompt set, and compare latency, token use, and errors. The goal is to learn whether routing outside the app makes the next backend change easier to test.