Streaming transforms the user experience by delivering tokens as they're generated. Instead of waiting 5-10 seconds for a complete response, users see the first token within 200-500ms — a dramatic improvement in perceived performance.
Key Benefits
- Time-to-first-token (TTFT) — Users see output almost immediately, reducing perceived latency by 80-90%.
- Progressive disclosure — Content unfolds naturally, like reading someone's thoughts in real-time.
- Cancel early — Users can stop generation if the response is going off-track, saving tokens and money.
- UI responsiveness — The interface feels alive and responsive rather than frozen.
Both the Chat Completions and Responses APIs support streaming, but they use different formats. Chat Completions uses flat data: lines with delta objects. The Responses API uses typed semantic events with event: and data: fields.