Streaming

A non-streaming call waits for the entire response to finish generating before you see anything. Streaming delivers tokens as they're produced, so a user sees text appearing immediately — the total generation time is the same, but the perceived latency is much lower.

with client.messages.stream(
    model="claude-sonnet-5", max_tokens=300,
    messages=[{"role": "user", "content": "Write a haiku about oceans."}],
) as stream:
    for text in stream.text_stream:
        print(text, end="", flush=True)

Under the hood this is usually server-sent events: a long-lived HTTP connection delivering a sequence of small chunks. Your code has to handle partial output correctly — a chunk isn't guaranteed to be a complete word, sentence, or JSON object, so code that assumes it can immediately parse each chunk as something complete will break.