MessagesKemampuan model

Streaming pesan

Lakukan streaming respons Messages API secara bertahap dengan server-sent events, termasuk delta teks, penggunaan alat, dan pemikiran diperpanjang.

Saat membuat sebuah Message, Anda dapat mengatur "stream": true untuk melakukan streaming respons secara bertahap menggunakan server-sent events (SSE).

Streaming dengan SDK

Python SDK dan TypeScript SDK menawarkan beberapa cara untuk melakukan streaming. PHP SDK menyediakan streaming melalui createStream(). Python SDK mendukung stream sinkron maupun asinkron. Lihat dokumentasi di masing-masing SDK untuk detailnya.

client = anthropic.Anthropic()

with client.messages.stream(
    max_tokens=1024,
    messages=[{"role": "user", "content": "Hello"}],
    model="claude-opus-5",
) as stream:
    for text in stream.text_stream:
        print(text, end="", flush=True)

Mendapatkan pesan akhir tanpa menangani event

Jika Anda tidak perlu memproses teks saat teks tersebut tiba, SDK menyediakan cara untuk menggunakan streaming secara internal sambil mengembalikan objek Message lengkap, identik dengan yang dikembalikan oleh .create(). Ini sangat berguna untuk permintaan dengan nilai max_tokens yang besar, di mana SDK mewajibkan streaming untuk menghindari timeout HTTP.

client = anthropic.Anthropic()

with client.messages.stream(
    max_tokens=128000,
    messages=[{"role": "user", "content": "Write a detailed analysis..."}],
    model="claude-opus-5",
) as stream:
    message = stream.get_final_message()

for block in message.content:
    if block.type == "text":
        print(block.text)

Pemanggilan .stream() menjaga koneksi HTTP tetap hidup dengan server-sent events, lalu .get_final_message() (Python) atau .finalMessage() (TypeScript) mengakumulasi semua event dan mengembalikan objek Message lengkap. Di Go, Anda memanggil message.Accumulate(event) di dalam loop stream untuk membangun Message lengkap yang sama. Di Java, gunakan MessageAccumulator.create() dan panggil accumulator.accumulate(event) pada setiap event. Di C#, lakukan await pada metode ekstensi .Aggregate() milik stream untuk mendapatkan Message lengkap, atau teruskan MessageContentAggregator ke .CollectAsync() untuk melakukan agregasi sambil menangani event. Di Ruby, panggil .accumulated_message pada stream. Di PHP SDK, Anda melakukan iterasi atas event stream secara manual untuk mengakumulasi respons.

Jenis event

Setiap server-sent event menyertakan jenis event bernama dan data JSON terkait. Setiap event menggunakan nama event SSE (misalnya, event: message_stop), dan menyertakan type event yang sesuai di dalam datanya.

Setiap stream menggunakan alur event berikut:

message_start: berisi objek Message dengan content kosong. Di bawah header beta thinking-binding-controls-2026-08-01, objek Message ini juga membawa array input_transformations. Setelah fallback sisi server di tengah stream, event message_delta terakhir membawa array tersebut lagi dengan entri dari model yang melayani.
Serangkaian blok konten, yang masing-masing memiliki event content_block_start, satu atau lebih event content_block_delta, dan event content_block_stop. Setiap blok konten memiliki index yang sesuai dengan indeksnya dalam array content Message akhir. Satu pengecualian: selama respons fallback sisi server, blok konten fallback tiba di setiap batas model sebagai pasangan content_block_start dan content_block_stop tanpa delta di antaranya.
Satu atau lebih event message_delta, yang menunjukkan perubahan tingkat atas pada objek Message akhir.
Event message_stop terakhir.

Event ping

Stream event juga dapat menyertakan sejumlah event ping.

Event error

API sesekali dapat mengirimkan error dalam stream event. Misalnya, selama periode penggunaan tinggi, Anda mungkin menerima overloaded_error, yang biasanya setara dengan HTTP 529 dalam konteks non-streaming:

Example error

event: error
data: {"type": "error", "error": {"type": "overloaded_error", "message": "Overloaded"}}

Event lainnya

Sesuai dengan kebijakan pembuatan versi, jenis event baru dapat ditambahkan, dan kode Anda harus menangani jenis event yang tidak dikenal dengan baik.

Jenis delta blok konten

Setiap event content_block_delta berisi delta dengan jenis yang memperbarui blok content pada index tertentu.

Delta teks

Delta blok konten text terlihat seperti:

Text delta

event: content_block_delta
data: {"type": "content_block_delta","index": 0,"delta": {"type": "text_delta", "text": "ello frien"}}

Delta JSON input

Delta untuk blok konten tool_use berkaitan dengan pembaruan pada field input dari blok tersebut. Untuk mendukung granularitas maksimum, delta berupa string JSON parsial, sedangkan tool_use.input akhir selalu berupa objek.

Anda dapat mengakumulasi delta string dan mem-parse JSON setelah menerima event content_block_stop, dengan menggunakan library seperti Pydantic untuk melakukan parsing JSON parsial, atau dengan menggunakan SDK, yang menyediakan helper untuk mengakses nilai inkremental yang telah di-parse.

Delta blok konten tool_use terlihat seperti:

Input JSON delta

event: content_block_delta
data: {"type": "content_block_delta","index": 1,"delta": {"type": "input_json_delta","partial_json": "{\"location\": \"San Fra"}}}

Catatan: Model saat ini hanya mendukung pengeluaran satu properti kunci dan nilai lengkap dari input dalam satu waktu. Oleh karena itu, saat menggunakan alat, mungkin ada jeda antara event streaming selagi model bekerja. Setelah kunci dan nilai input terakumulasi, keduanya dikeluarkan sebagai beberapa event content_block_delta dengan JSON parsial yang dipotong-potong sehingga format ini dapat secara otomatis mendukung granularitas yang lebih halus pada model mendatang.

Delta thinking

Saat menggunakan thinking dengan streaming diaktifkan, Anda akan menerima konten thinking melalui event thinking_delta. Delta ini berkaitan dengan field thinking dari blok konten thinking.

Untuk konten thinking, event signature_delta khusus dikirim tepat sebelum event content_block_stop. Signature ini digunakan untuk memverifikasi integritas blok thinking.

Ketika display: "omitted" diatur pada konfigurasi thinking, tidak ada event thinking_delta yang dikirim. Blok thinking dibuka, menerima satu signature_delta, lalu ditutup. Dengan display: "updates" (beta), blok penalaran di-stream dengan cara yang sama, dan hanya pembaruan progres yang ditulis beberapa model di antara pemanggilan alat yang men-stream event thinking_delta. Lihat Mengontrol tampilan thinking.

Delta thinking yang umum terlihat seperti:

Thinking delta

event: content_block_delta
data: {"type": "content_block_delta", "index": 0, "delta": {"type": "thinking_delta", "thinking": "I need to find the GCD of 1071 and 462 using the Euclidean algorithm.\n\n1071 = 2 × 462 + 147"}}

Delta signature terlihat seperti:

Signature delta

event: content_block_delta
data: {"type": "content_block_delta", "index": 0, "delta": {"type": "signature_delta", "signature": "EqQBCgIYAhIM1gbcDa9GJwZA2b3hGgxBdjrkzLoky3dl1pkiMOYds..."}}

Respons stream HTTP lengkap

Gunakan SDK klien saat menggunakan mode streaming. Namun, jika Anda membangun integrasi API langsung, Anda perlu menangani event ini sendiri.

Respons stream terdiri dari:

Event message_start
Kemungkinan beberapa blok konten, yang masing-masing berisi:
- Event content_block_start
- Kemungkinan beberapa event content_block_delta
- Event content_block_stop
Satu atau lebih event message_delta
Event message_stop

Mungkin juga terdapat event ping yang tersebar di sepanjang respons. Lihat Jenis event untuk detail lebih lanjut tentang formatnya.

Permintaan streaming dasar

client = anthropic.Anthropic()

with client.messages.stream(
    model="claude-opus-5",
    messages=[{"role": "user", "content": "Hello"}],
    max_tokens=256,
) as stream:
    for text in stream.text_stream:
        print(text, end="", flush=True)

Response

event: message_start
data: {"type": "message_start", "message": {"id": "msg_1nZdL29xx5MUA1yADyHTEsnR8uuvGzszyY", "type": "message", "role": "assistant", "content": [], "model": "claude-opus-5", "stop_reason": null, "stop_sequence": null, "usage": {"input_tokens": 25, "output_tokens": 1}}}

event: content_block_start
data: {"type": "content_block_start", "index": 0, "content_block": {"type": "text", "text": ""}}

event: ping
data: {"type": "ping"}

event: content_block_delta
data: {"type": "content_block_delta", "index": 0, "delta": {"type": "text_delta", "text": "Hello"}}

event: content_block_delta
data: {"type": "content_block_delta", "index": 0, "delta": {"type": "text_delta", "text": "!"}}

event: content_block_stop
data: {"type": "content_block_stop", "index": 0}

event: message_delta
data: {"type": "message_delta", "delta": {"stop_reason": "end_turn", "stop_sequence":null}, "usage": {"output_tokens": 15}}

event: message_stop
data: {"type": "message_stop"}

Permintaan streaming dengan penggunaan alat

Permintaan ini meminta Claude menggunakan alat untuk melaporkan cuaca.

client = anthropic.Anthropic()

tools = [
    {
        "name": "get_weather",
        "description": "Get the current weather in a given location",
        "input_schema": {
            "type": "object",
            "properties": {
                "location": {
                    "type": "string",
                    "description": "The city and state, e.g. San Francisco, CA",
                }
            },
            "required": ["location"],
        },
    }
]

with client.messages.stream(
    model="claude-opus-5",
    max_tokens=1024,
    tools=tools,
    tool_choice={"type": "any"},
    messages=[
        {"role": "user", "content": "What is the weather like in San Francisco?"}
    ],
) as stream:
    for text in stream.text_stream:
        print(text, end="", flush=True)

Response

event: message_start
data: {"type":"message_start","message":{"id":"msg_014p7gG3wDgGV9EUtLvnow3U","type":"message","role":"assistant","model":"claude-opus-5","stop_sequence":null,"usage":{"input_tokens":472,"output_tokens":2},"content":[],"stop_reason":null}}

event: content_block_start
data: {"type":"content_block_start","index":0,"content_block":{"type":"text","text":""}}

event: ping
data: {"type": "ping"}

event: content_block_delta
data: {"type":"content_block_delta","index":0,"delta":{"type":"text_delta","text":"Okay"}}

event: content_block_delta
data: {"type":"content_block_delta","index":0,"delta":{"type":"text_delta","text":","}}

event: content_block_delta
data: {"type":"content_block_delta","index":0,"delta":{"type":"text_delta","text":" let"}}

event: content_block_delta
data: {"type":"content_block_delta","index":0,"delta":{"type":"text_delta","text":"'s"}}

event: content_block_delta
data: {"type":"content_block_delta","index":0,"delta":{"type":"text_delta","text":" check"}}

event: content_block_delta
data: {"type":"content_block_delta","index":0,"delta":{"type":"text_delta","text":" the"}}

event: content_block_delta
data: {"type":"content_block_delta","index":0,"delta":{"type":"text_delta","text":" weather"}}

event: content_block_delta
data: {"type":"content_block_delta","index":0,"delta":{"type":"text_delta","text":" for"}}

event: content_block_delta
data: {"type":"content_block_delta","index":0,"delta":{"type":"text_delta","text":" San"}}

event: content_block_delta
data: {"type":"content_block_delta","index":0,"delta":{"type":"text_delta","text":" Francisco"}}

event: content_block_delta
data: {"type":"content_block_delta","index":0,"delta":{"type":"text_delta","text":","}}

event: content_block_delta
data: {"type":"content_block_delta","index":0,"delta":{"type":"text_delta","text":" CA"}}

event: content_block_delta
data: {"type":"content_block_delta","index":0,"delta":{"type":"text_delta","text":":"}}

event: content_block_stop
data: {"type":"content_block_stop","index":0}

event: content_block_start
data: {"type":"content_block_start","index":1,"content_block":{"type":"tool_use","id":"toolu_01T1x1fJ34qAmk2tNTrN7Up6","name":"get_weather","input":{}}}

event: content_block_delta
data: {"type":"content_block_delta","index":1,"delta":{"type":"input_json_delta","partial_json":""}}

event: content_block_delta
data: {"type":"content_block_delta","index":1,"delta":{"type":"input_json_delta","partial_json":"{\"location\":"}}

event: content_block_delta
data: {"type":"content_block_delta","index":1,"delta":{"type":"input_json_delta","partial_json":" \"San"}}

event: content_block_delta
data: {"type":"content_block_delta","index":1,"delta":{"type":"input_json_delta","partial_json":" Francisc"}}

event: content_block_delta
data: {"type":"content_block_delta","index":1,"delta":{"type":"input_json_delta","partial_json":"o,"}}

event: content_block_delta
data: {"type":"content_block_delta","index":1,"delta":{"type":"input_json_delta","partial_json":" CA\"}"}}

event: content_block_stop
data: {"type":"content_block_stop","index":1}

event: message_delta
data: {"type":"message_delta","delta":{"stop_reason":"tool_use","stop_sequence":null},"usage":{"output_tokens":89}}

event: message_stop
data: {"type":"message_stop"}

Permintaan streaming dengan thinking

Permintaan ini mengaktifkan thinking dengan streaming. Pengaturan display: "summarized" men-stream ringkasan padat dari penalaran Claude alih-alih rantai pemikiran lengkap.

client = anthropic.Anthropic()

with client.messages.stream(
    model="claude-opus-5",
    max_tokens=20000,
    thinking={"type": "adaptive", "display": "summarized"},
    messages=[
        {
            "role": "user",
            "content": "What is the greatest common divisor of 1071 and 462?",
        }
    ],
) as stream:
    for event in stream:
        if event.type == "content_block_delta":
            if event.delta.type == "thinking_delta":
                print(event.delta.thinking, end="", flush=True)
            elif event.delta.type == "text_delta":
                print(event.delta.text, end="", flush=True)

Response

event: message_start
data: {"type": "message_start", "message": {"id": "msg_01...", "type": "message", "role": "assistant", "content": [], "model": "claude-opus-5", "stop_reason": null, "stop_sequence": null}}

event: content_block_start
data: {"type": "content_block_start", "index": 0, "content_block": {"type": "thinking", "thinking": "", "signature": ""}}

event: content_block_delta
data: {"type": "content_block_delta", "index": 0, "delta": {"type": "thinking_delta", "thinking": "I need to find the GCD of 1071 and 462 using the Euclidean algorithm.\n\n1071 = 2 × 462 + 147"}}

event: content_block_delta
data: {"type": "content_block_delta", "index": 0, "delta": {"type": "thinking_delta", "thinking": "\n462 = 3 × 147 + 21"}}

event: content_block_delta
data: {"type": "content_block_delta", "index": 0, "delta": {"type": "thinking_delta", "thinking": "\n147 = 7 × 21 + 0"}}

event: content_block_delta
data: {"type": "content_block_delta", "index": 0, "delta": {"type": "thinking_delta", "thinking": "\nThe remainder is 0, so GCD(1071, 462) = 21."}}

event: content_block_delta
data: {"type": "content_block_delta", "index": 0, "delta": {"type": "signature_delta", "signature": "EqQBCgIYAhIM1gbcDa9GJwZA2b3hGgxBdjrkzLoky3dl1pkiMOYds..."}}

event: content_block_stop
data: {"type": "content_block_stop", "index": 0}

event: content_block_start
data: {"type": "content_block_start", "index": 1, "content_block": {"type": "text", "text": ""}}

event: content_block_delta
data: {"type": "content_block_delta", "index": 1, "delta": {"type": "text_delta", "text": "The greatest common divisor of 1071 and 462 is **21**."}}

event: content_block_stop
data: {"type": "content_block_stop", "index": 1}

event: message_delta
data: {"type": "message_delta", "delta": {"stop_reason": "end_turn", "stop_sequence": null}}

event: message_stop
data: {"type": "message_stop"}

Permintaan streaming dengan penggunaan alat web search

Permintaan ini meminta Claude mencari informasi cuaca terkini di web.

client = anthropic.Anthropic()

with client.messages.stream(
    model="claude-opus-5",
    max_tokens=1024,
    tools=[{"type": "web_search_20250305", "name": "web_search", "max_uses": 5}],
    messages=[
        {"role": "user", "content": "What is the weather like in New York City today?"}
    ],
) as stream:
    for text in stream.text_stream:
        print(text, end="", flush=True)

Response

event: message_start
data: {"type":"message_start","message":{"id":"msg_01G...","type":"message","role":"assistant","model":"claude-opus-5","content":[],"stop_reason":null,"stop_sequence":null,"usage":{"input_tokens":2679,"cache_creation_input_tokens":0,"cache_read_input_tokens":0,"output_tokens":3}}}

event: content_block_start
data: {"type":"content_block_start","index":0,"content_block":{"type":"text","text":""}}

event: content_block_delta
data: {"type":"content_block_delta","index":0,"delta":{"type":"text_delta","text":"I'll check"}}

event: content_block_delta
data: {"type":"content_block_delta","index":0,"delta":{"type":"text_delta","text":" the current weather in New York City for you"}}

event: ping
data: {"type": "ping"}

event: content_block_delta
data: {"type":"content_block_delta","index":0,"delta":{"type":"text_delta","text":"."}}

event: content_block_stop
data: {"type":"content_block_stop","index":0}

event: content_block_start
data: {"type":"content_block_start","index":1,"content_block":{"type":"server_tool_use","id":"srvtoolu_014hJH82Qum7Td6UV8gDXThB","name":"web_search","input":{}}}

event: content_block_delta
data: {"type":"content_block_delta","index":1,"delta":{"type":"input_json_delta","partial_json":""}}

event: content_block_delta
data: {"type":"content_block_delta","index":1,"delta":{"type":"input_json_delta","partial_json":"{\"query"}}

event: content_block_delta
data: {"type":"content_block_delta","index":1,"delta":{"type":"input_json_delta","partial_json":"\":"}}

event: content_block_delta
data: {"type":"content_block_delta","index":1,"delta":{"type":"input_json_delta","partial_json":" \"weather"}}

event: content_block_delta
data: {"type":"content_block_delta","index":1,"delta":{"type":"input_json_delta","partial_json":" NY"}}

event: content_block_delta
data: {"type":"content_block_delta","index":1,"delta":{"type":"input_json_delta","partial_json":"C to"}}

event: content_block_delta
data: {"type":"content_block_delta","index":1,"delta":{"type":"input_json_delta","partial_json":"day\"}"}}

event: content_block_stop
data: {"type":"content_block_stop","index":1 }

event: content_block_start
data: {"type":"content_block_start","index":2,"content_block":{"type":"web_search_tool_result","tool_use_id":"srvtoolu_014hJH82Qum7Td6UV8gDXThB","content":[{"type":"web_search_result","title":"Weather in New York City in May 2025 (New York) - detailed Weather Forecast for a month","url":"https://world-weather.info/forecast/usa/new_york/may-2025/","encrypted_content":"Ev0DCioIAxgCIiQ3NmU4ZmI4OC1k...","page_age":null},...]}}

event: content_block_stop
data: {"type":"content_block_stop","index":2}

event: content_block_start
data: {"type":"content_block_start","index":3,"content_block":{"type":"text","text":""}}

event: content_block_delta
data: {"type":"content_block_delta","index":3,"delta":{"type":"text_delta","text":"Here's the current weather information for New York"}}

event: content_block_delta
data: {"type":"content_block_delta","index":3,"delta":{"type":"text_delta","text":" City:\n\n# Weather"}}

event: content_block_delta
data: {"type":"content_block_delta","index":3,"delta":{"type":"text_delta","text":" in New York City"}}

event: content_block_delta
data: {"type":"content_block_delta","index":3,"delta":{"type":"text_delta","text":"\n\n"}}

...

event: content_block_stop
data: {"type":"content_block_stop","index":17}

event: message_delta
data: {"type":"message_delta","delta":{"stop_reason":"end_turn","stop_sequence":null},"usage":{"input_tokens":10682,"cache_creation_input_tokens":0,"cache_read_input_tokens":0,"output_tokens":510,"server_tool_use":{"web_search_requests":1}}}

event: message_stop
data: {"type":"message_stop"}

Pemulihan error

Claude 4.5 dan sebelumnya

Untuk model Claude 4.5 dan sebelumnya, Anda dapat memulihkan permintaan streaming yang terputus karena masalah jaringan, timeout, atau error lainnya dengan melanjutkan dari titik stream terputus. Pendekatan ini menghindarkan Anda dari pemrosesan ulang seluruh respons.

Strategi pemulihan dasar meliputi:

Tangkap respons parsial: Simpan semua konten yang berhasil diterima sebelum error terjadi.
Susun permintaan lanjutan: Buat permintaan API baru yang menyertakan respons asisten parsial sebagai awal dari pesan asisten baru.
Lanjutkan streaming: Lanjutkan menerima sisa respons dari titik terputusnya.

Claude 4.6 dan setelahnya

Untuk model Claude 4.6 dan setelahnya, strategi tangkap-dan-lanjutkan yang sama berlaku, tetapi langkah 2 berubah: alih-alih menempatkan respons parsial dalam pesan asisten, tambahkan pesan pengguna yang menginstruksikan model untuk melanjutkan dari titik terakhirnya.

Tangkap respons parsial: Simpan semua konten yang berhasil diterima sebelum error terjadi.
Susun permintaan lanjutan: Buat permintaan API baru dengan pesan pengguna yang berisi respons parsial dan instruksi untuk melanjutkan, misalnya:
Sample prompt
```
Your previous response was interrupted and ended with [previous_response]. Continue from where you left off.
```
Lanjutkan streaming: Lanjutkan menerima sisa respons dari titik terputusnya.

Praktik terbaik pemulihan error

Gunakan fitur SDK: Manfaatkan kemampuan akumulasi pesan dan penanganan error bawaan SDK.
Tangani jenis konten: Perhatikan bahwa pesan dapat berisi beberapa blok konten (text, tool_use, thinking). Blok penggunaan alat dan pemikiran diperpanjang tidak dapat dipulihkan sebagian. Anda dapat melanjutkan streaming dari blok teks terbaru.

Langkah selanjutnya

Alasan berhenti dan fallback

Tangani setiap nilai stop_reason setelah stream selesai.

Streaming alat fine-grained

Stream JSON input alat tanpa buffering sisi server untuk latensi lebih rendah.

Thinking

Stream output thinking dengan event thinking_delta dan signature_delta.

SDK klien

Gunakan SDK resmi, yang menangani streaming, akumulasi, dan koneksi ulang untuk Anda.

Pemrosesan batch

Proses permintaan dalam volume besar secara asinkron ketika Anda tidak memerlukan respons real-time.

Streaming dengan SDK

Mendapatkan pesan akhir tanpa menangani event

Jenis event

Event ping

Event error

Event lainnya

Jenis delta blok konten

Delta teks

Delta JSON input

Delta thinking

Respons stream HTTP lengkap

Permintaan streaming dasar

Permintaan streaming dengan penggunaan alat

Permintaan streaming dengan thinking

Permintaan streaming dengan penggunaan alat web search

Pemulihan error

Claude 4.5 dan sebelumnya

Claude 4.6 dan setelahnya

Praktik terbaik pemulihan error

Langkah selanjutnya

Streaming dengan SDK

Mendapatkan pesan akhir tanpa menangani event

Jenis event

Event ping

Event error

Event lainnya

Jenis delta blok konten

Delta teks

Delta JSON input

Delta thinking

Respons stream HTTP lengkap

Permintaan streaming dasar

Permintaan streaming dengan penggunaan alat

Permintaan streaming dengan thinking

Permintaan streaming dengan penggunaan alat web search

Pemulihan error

Claude 4.5 dan sebelumnya

Claude 4.6 dan setelahnya

Praktik terbaik pemulihan error

Langkah selanjutnya