Question

A development team is building a customer-facing chat assistant that invokes an Anthropic Claude model on Amazon Bedrock. Product requirements state that partial responses must appear in the browser token-by-token as the model generates them, to reduce perceived latency. The front end is a single-page web app, and the backend uses AWS Lambda invoked through Amazon API Gateway. Which architecture correctly delivers the streaming experience end-to-end?

Accepted Answer

Use API Gateway WebSocket API with a Lambda integration that calls InvokeModelWithResponseStream and pushes each chunk back to the client via the connection's callback URL.

Answer

Use API Gateway REST API with a Lambda proxy integration that calls InvokeModel and returns the full completion once the model finishes.

Answer

Use API Gateway HTTP API with a Lambda integration that calls InvokeModelWithResponseStream and returns the accumulated chunks as a single JSON array in the HTTP response.

Answer

Enable Bedrock provisioned throughput and let the SPA call the Bedrock runtime endpoint directly so the browser reads the streamed chunks without any backend.