As reported on November 17, 2026, Google announced a public preview of streaming support for Cloud API Gateway. The change is aimed in part at AI applications that deliver model output gradually or provide progress updates during multi-step tasks.
Google says streaming endpoints can now run behind API Gateway while retaining existing controls, including API key enforcement, JWT validation, backend service-account authentication, Cloud Logging and Cloud Monitoring integration, and request routing. Previously, developers had to choose between using the gateway without streaming or routing streaming traffic around it—and losing those gateway controls.
The preview supports server-sent events (SSE), chunked transfer encoding and HTTP response streaming, WebSockets, and gRPC. A streaming connection can remain open for up to one hour; clients must establish a new connection to continue after that. Google says existing API specifications need no changes or special annotations. Developers must enable streaming when creating a gateway using the --enable-streaming flag.
The release also introduces predictable gateway hostnames in the format GATEWAY_ID-PROJECT_NUMBER.REGION.gateway.dev. Unlike the previous hash-based hostnames, these can be known before a gateway is created. Google says its model-routing feature can provide a single OpenAI-compatible endpoint that routes requests to Gemini, Claude or OpenAI gpt-oss models according to the model field in the request.
There are preview limitations to consider. X user emildrops noted that existing gateways cannot be switched to streaming in place. Both emildrops and LinkedIn user Ehsan Elahi Zaheer warned that the gateway’s request-latency metric measures the duration of the full stream, not time to first byte, so a latency chart may not reflect how quickly a user first sees a response.


