Your chat assistant's long answers take 8-15 seconds to finish generating. Analytics show many users abandon the page before the response appears, even though answer quality ratings are high. What change best fixes perceived latency without hurting quality?

1. A Generate the full response server-side, then return it in one JSON payload with a progress spinner so clients render it atomically.
2. B Have the client poll a status endpoint every two seconds and fetch the completed response once generation finishes on the server.
3. C Stream tokens to the client over SSE as they are generated, persisting the final message once the stream has completed.
4. D Switch every chat request to the smallest available model so full responses finish in under a second, regardless of quality.