Infrastructure & Compute
Decode phase
The stage when a language AI builds its reply, one small piece at a time.
Example
An answer appears in small pieces after the AI has read the request.
Why people use it
It helps explain the part of the wait spent producing the answer.
What you'll hear
“It has read the request and is now writing the reply.”
What this means for you
Measure both stages when diagnosing slow responses.
Can you control it?
Developer-only
The people building or running the AI choose this setup. An everyday user generally needs their help to change how this part works.
Common questions
- Does this stage include reading the original request?
- Reading and preparing the request is usually treated as an earlier stage; this stage builds the reply.
- Can a longer answer take longer at this stage?
- Yes. Producing more pieces of the reply generally adds work.
- Can the answer arrive gradually?
- Yes. Many services display pieces as they are produced instead of waiting for the whole reply.