Streaming
Also known as: stream · token streaming
Streaming is delivering the model’s output as it’s generated, token by token, instead of waiting for the full response.
The user sees the answer build up live — dramatically better perceived speed.
- Without streaming: wait 8s, then the whole answer appears.
- With streaming: first words appear in ~1s, then it flows.
Always stream for UX: even if total time is identical, "it's already writing" feels 3× faster than a blank screen. It's the single biggest perceived-performance win in LLM apps.