Streaming means the answer arrives while the model is still writing it, piece by piece instead of all at once. It can make the model feel faster even though the full run takes the same time: the early pieces arrive as soon as they’re ready, which lowers the felt latency. A pause mid-answer doesn’t always mean the system is stuck; the model may be waiting on a tool call to finish before it continues.
Imagine watching someone type a reply in a shared document. You don’t wait for them to print the final page and hand it over. You see the first sentence, then the next. The work is still in progress, and you’re already reading.
Why you care
Streaming lets you supervise earlier. You can catch tone, direction, and misunderstandings before the full answer lands: if you asked for a client email and the first paragraph is clearly wrong, you can stop it there. A streamed answer can still be wrong; the work just becomes visible sooner. Seeing work sooner means steering sooner, and steering early is usually cheaper than cleaning up late.