Google Cloud has added model routing to API Gateway in public preview, giving developers a serverless way to send OpenAI-compatible AI requests to different model backends without managing a separate proxy layer.

The feature lets teams define routing rules inside an OpenAPI 3.x specification. Instead of hardcoding every client to a provider-specific endpoint, developers can map virtual model names to backend targets and let API Gateway act as the ingress point. Google's example shows routing requests across Gemini, Anthropic Claude, and OpenAI OSS-GPT endpoints hosted through Google Cloud paths.

That makes the release less about a new model and more about where model-selection logic lives. Many production AI applications now call several model families for cost, latency, quality, fallback, or policy reasons. Putting those choices behind API Gateway could make it easier for platform teams to centralize authentication, endpoint definitions, and traffic rules while keeping application code pointed at a stable interface.

The practical value will depend on how teams already use Google Cloud API Gateway and whether the preview feature covers their provider mix, observability needs, and failure behavior. It also does not remove the harder work of evaluating which model should handle a given request.

Still, the move is another sign that model routing is becoming a normal developer-infrastructure concern. Cloud gateways, AI gateways, and application frameworks are all competing to own the layer between AI apps and the fast-changing model catalog beneath them.