Streaming

Add stream: true to a request for any model to receive incremental SSE (Server-Sent Events) output and reduce the wait for the first token.

OpenAI protocol

stream = client.chat.completions.create(
    model="z-ai/glm-5.2",
    messages=[{"role": "user", "content": "Write a short poem about computing power"}],
    stream=True,
)
for chunk in stream:
    delta = chunk.choices[0].delta.content
    if delta:
        print(delta, end="")

Reasoning models return incremental reasoning in delta.reasoning_content and answer text in delta.content.

Get usage in a stream

Streaming responses do not include usage by default. Add stream_options to receive complete usage in the final data chunk, whose choices is an empty array:

{"stream": true, "stream_options": {"include_usage": true}}

Anthropic protocol

The Messages endpoint follows the Anthropic SSE sequence: message_start → content_block_start → content_block_delta (text_delta / thinking_delta / input_json_delta) → content_block_stop → message_delta (including stop_reason and output token counts) → message_stop.

Use client.messages.stream(...) from the official Anthropic SDK directly.

Recommendations

• Enable streaming for long outputs and complex reasoning to avoid client or intermediate proxy read timeouts.
• Streaming and non-streaming requests are billed the same way, using the number of tokens actually generated.
• An upstream error during generation terminates the stream. Treat early termination without finish_reason / message_stop as a failure and retry.