Streaming
Set stream to true on a chat request and the answer comes back as server-sent events, in the same shape OpenAI uses.
What you are charged
The final chunk carries usage with prompt and completion tokens. A keyed request is charged from that count, so the console matches your logs to the token.
A paid x402 request streams the same events. Its USDC amount is fixed before the first byte, from the prompt and the output cap, and does not change if the answer is shorter.
Decisions are not streamed: a decisions call answers in one JSON body.