Streaming
Sending model output to the user token-by-token as it's generated, rather than waiting for the full response before rendering.
Streaming is the standard UX pattern for modern AI chat products. Instead of waiting five seconds for a 500-token response and then rendering the whole thing, streaming shows each token as it's generated — you read the first sentence while the model is still writing the fourth.
The technical reality: the model generates tokens sequentially anyway. Non-streaming responses aren't faster; they're just held back until complete. Streaming is a pure UX win — same underlying speed, but the perceived latency drops dramatically because "time to first useful character" replaces "time to full response".
The one honest caveat: streaming can expose you to partial hallucinations. If a response starts confidently down a wrong path and gets corrected in the last paragraph, you might act on the wrong middle before the correction lands. In practice this is rare with well-trained models, and the ability to interrupt a bad response early usually compensates.