Streaming

Set stream: true and tokens arrive as Server-Sent Events. The protocol is the same as OpenAI's, so an existing stream handler works unchanged.

Stream a response#

stream = client.chat.completions.create(
    model="deepseek/deepseek-v4-pro",
    messages=[{"role": "user", "content": "Count to ten."}],
    stream=True,
)

for event in stream:
    delta = event.choices[0].delta.content
    if delta:
        print(delta, end="", flush=True)

The stream ends with a data: [DONE] line, exactly like OpenAI.

Get token counts while streaming#

Add "stream_options": {"include_usage": true} to the request body to get a final chunk with the token counts for the whole request.

stream = client.chat.completions.create(
    model="deepseek/deepseek-v4-pro",
    messages=[{"role": "user", "content": "Count to ten."}],
    stream=True,
    stream_options={"include_usage": True},
)

Streaming and spend limits#

Budgets are checked when a request starts, not while it runs. A stream in progress is never cut off, so a request that starts just under a limit can finish and push usage slightly over it. The next request is refused.

Reasoning models#

Models that return their thinking send it separately from the answer in the stream. See Reasoning & tracing.