Streaming

Receive model output incrementally

Add "stream": true to supported JSON requests. Chat Completions and Responses generally return Server-Sent Events. Read each data: record and handle the protocol’s terminal event.

{"model":"your-model","messages":[{"role":"user","content":"Hello"}],"stream":true}

A disconnected request may still incur usage. Apply exponential backoff and do not blindly replay non-idempotent requests.