AnyoneLearnAI
Start

Reference · API

Anthropic streaming messages

Last updated

Stream Claude replies **token-by-token** for faster perceived latency in chat UIs.

Stream Claude replies **token-by-token** for faster perceived latency in chat UIs.

#When to use

  • Interactive assistants where users wait on screen
  • Long answers (summaries, code explanations)
  • Server-sent events (SSE) or WebSocket frontends

    • Use non-streaming when you need the full response atomically for parsing or tool routing.

    #Endpoint

    POST https://api.anthropic.com/v1/messages

    Set `"stream": true` in the JSON body (or use SDK stream helpers).

    #Request shape (sketch)

    {
  "model": "claude-sonnet-4-20250514",
  "max_tokens": 1024,
  "stream": true,
  "messages": [{"role": "user", "content": "Explain RAG briefly."}]
}

    #Python sketch

    with client.messages.stream(
    model="claude-sonnet-4-20250514",
    max_tokens=512,
    messages=[{"role": "user", "content": "Hello"}],
) as stream:
    for text in stream.text_stream:
        print(text, end="", flush=True)

    #Event types

    Stream emits `message_start`, `content_block_delta`, `message_delta`, `message_stop` — handle partial JSON if you parse tool use mid-stream.

    #Common pitfalls

    | Pitfall | Fix |

    |---------|-----|

    | Dropping partial tokens on error | Flush buffers; show retry UI |

    | Parsing tools before stream ends | Wait for `message_stop` or use SDK helpers |

    | Missing `max_tokens` | Required on every Anthropic request |

    | Streaming from browser with raw key | Proxy through your server |

    Learn by doing → Try the interactive lesson →
    ← All API docs