Streaming (Token Streaming)

Sending partial model outputs as they are generated to reduce perceived latency.