ChatBedrock

This doc will help you get started with AWS Bedrock chat models. Amazon Bedrock is a fully managed service that offers a choice of high-performing foundation models (FMs) from leading AI companies like AI21 Labs, Anthropic, Cohere, Meta, Stability AI, and Amazon via a single API, along with a broad set of capabilities you need to build generative AI applications with security, privacy, and responsible AI. Using Amazon Bedrock, you can easily experiment with and evaluate top FMs for your use case, privately customize them with your data using techniques such as fine-tuning and Retrieval Augmented Generation (RAG), and build agents that execute tasks using your enterprise systems and data sources. Since Amazon Bedrock is serverless, you don’t have to manage any infrastructure, and you can securely integrate and deploy generative AI capabilities into your applications using the AWS services you are already familiar with.

AWS Bedrock maintains a Converse API which provides a unified conversational interface for Bedrock models. This API does not yet support custom models. You can see a list of all models that are supported here.

We recommend the Converse API for users who do not need to use custom models. It can be accessed using ChatBedrockConverse.

For detailed documentation of all Bedrock features and configurations head to the API reference.

Overview

Integration details

Class	Package	Serializable	JS support	Downloads	Version
ChatBedrock	langchain-aws	beta	✅
ChatBedrockConverse	langchain-aws	beta	✅

Model features

The below apply to both ChatBedrock and ChatBedrockConverse.

Tool calling	Structured output	Image input	Audio input	Video input	Token-level streaming	Native async	Token usage	Logprobs
✅	✅	✅	❌	❌	✅	❌	✅	❌

Setup

To access Bedrock models you’ll need to create an AWS account, set up the Bedrock API service, get an access key ID and secret key, and install the langchain-aws integration package.

Credentials

Head to the AWS docs to sign up to AWS and setup your credentials.

Alternatively, ChatBedrockConverse will read from the following environment variables by default:

# os.environ["AWS_ACCESS_KEY_ID"] = "..."
# os.environ["AWS_SECRET_ACCESS_KEY"] = "..."

# Not required unless using temporary credentials.
# os.environ["AWS_SESSION_TOKEN"] = "..."

You’ll also need to turn on model access for your account, which you can do by following these instructions.

To enable automated tracing of your model calls, set your LangSmith API key:

os.environ["LANGSMITH_API_KEY"] = getpass.getpass("Enter your LangSmith API key: ")
os.environ["LANGSMITH_TRACING"] = "true"

Installation

The LangChain Bedrock integration lives in the langchain-aws package:

pip install -qU langchain-aws

Instantiation

Now we can instantiate our model object and generate chat completions:

from langchain_aws import ChatBedrockConverse

llm = ChatBedrockConverse(
    model_id="anthropic.claude-3-5-sonnet-20240620-v1:0",
    # region_name=...,
    # aws_access_key_id=...,
    # aws_secret_access_key=...,
    # aws_session_token=...,
    # temperature=...,
    # max_tokens=...,
    # other params...
)

Invocation

messages = [
    (
        "system",
        "You are a helpful assistant that translates English to French. Translate the user sentence.",
    ),
    ("human", "I love programming."),
]
ai_msg = llm.invoke(messages)
ai_msg

AIMessage(content="J'adore la programmation.", additional_kwargs={}, response_metadata={'ResponseMetadata': {'RequestId': 'b07d1630-06f2-44b1-82bf-e82538dd2215', 'HTTPStatusCode': 200, 'HTTPHeaders': {'date': 'Wed, 16 Apr 2025 19:35:34 GMT', 'content-type': 'application/json', 'content-length': '206', 'connection': 'keep-alive', 'x-amzn-requestid': 'b07d1630-06f2-44b1-82bf-e82538dd2215'}, 'RetryAttempts': 0}, 'stopReason': 'end_turn', 'metrics': {'latencyMs': [488]}, 'model_name': 'anthropic.claude-3-5-sonnet-20240620-v1:0'}, id='run-d09ed928-146a-4336-b1fd-b63c9e623494-0', usage_metadata={'input_tokens': 29, 'output_tokens': 11, 'total_tokens': 40, 'input_token_details': {'cache_creation': 0, 'cache_read': 0}})

print(ai_msg.content)

J'adore la programmation.

Streaming

Note that ChatBedrockConverse emits content blocks while streaming:

for chunk in llm.stream(messages):
    print(chunk)

content=[] additional_kwargs={} response_metadata={} id='run-d0e0836e-7146-4c3d-97c7-ad23dac6febd'
content=[{'type': 'text', 'text': 'J', 'index': 0}] additional_kwargs={} response_metadata={} id='run-d0e0836e-7146-4c3d-97c7-ad23dac6febd'
content=[{'type': 'text', 'text': "'adore la", 'index': 0}] additional_kwargs={} response_metadata={} id='run-d0e0836e-7146-4c3d-97c7-ad23dac6febd'
content=[{'type': 'text', 'text': ' programmation.', 'index': 0}] additional_kwargs={} response_metadata={} id='run-d0e0836e-7146-4c3d-97c7-ad23dac6febd'
content=[{'index': 0}] additional_kwargs={} response_metadata={} id='run-d0e0836e-7146-4c3d-97c7-ad23dac6febd'
content=[] additional_kwargs={} response_metadata={'stopReason': 'end_turn'} id='run-d0e0836e-7146-4c3d-97c7-ad23dac6febd'
content=[] additional_kwargs={} response_metadata={'metrics': {'latencyMs': 600}, 'model_name': 'anthropic.claude-3-5-sonnet-20240620-v1:0'} id='run-d0e0836e-7146-4c3d-97c7-ad23dac6febd' usage_metadata={'input_tokens': 29, 'output_tokens': 11, 'total_tokens': 40, 'input_token_details': {'cache_creation': 0, 'cache_read': 0}}

You can filter to text using the .text property on the output:

for chunk in llm.stream(messages):
    print(chunk.text, end="|")

|J|'adore la| programmation.||||

Streaming tool calls and structured output

When using tool calling or structured output with Anthropic models, tool call arguments stream as partial JSON chunks by default.

To reduce latency and get more evenly distributed chunks, you can enable Anthropic’s fine-grained tool streaming beta:

from langchain_aws import ChatBedrockConverse

llm = ChatBedrockConverse(
    model_id="us.anthropic.claude-sonnet-4-5-20250514-v1:0",
    additional_model_request_fields={
        "anthropic_beta": ["fine-grained-tool-streaming-2025-05-14"]
    }
)

Fine-grained tool streaming is supported on Claude 4.5+ models. See the Claude documentation for more details.

When using fine-grained tool streaming, you may receive invalid or partial JSON inputs. Make sure to account for these edge cases in your code.

Extended thinking

This guide focuses on implementing Extended Thinking using AWS Bedrock with LangChain’s ChatBedrockConverse integration.

Supported models

Extended Thinking is available for the following Claude models on AWS Bedrock:

Model	Model ID
Claude Opus 4	anthropic.claude-opus-4-20250514-v1:0
Claude Sonnet 4	anthropic.claude-sonnet-4-20250514-v1:0
Claude 3.7 Sonnet	us.anthropic.claude-3-7-sonnet-20250219-v1:0

from langchain_aws import ChatBedrockConverse

llm = ChatBedrockConverse(
    model_id="us.anthropic.claude-sonnet-4-20250514-v1:0",
    region_name="us-west-2",
    max_tokens=4096,
    additional_model_request_fields={
        "thinking": {"type": "enabled", "budget_tokens": 1024},
    },
)

ai_msg = llm.invoke(messages)
ai_msg.content_blocks

[{'type': 'reasoning',
  'reasoning': 'The user wants me to translate "I love programming" from English to French.\n\n"I love" translates to "J\'aime" in French.\n"Programming" translates to "la programmation" in French.\n\nSo the full translation would be "J\'aime la programmation."',
  'extras': {'signature': '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'}},
 {'type': 'text', 'text': "J'aime la programmation."}]

How extended thinking works

When extended thinking is turned on, Claude creates thinking content blocks where it outputs its internal reasoning. Claude incorporates insights from this reasoning before crafting a final response. The API response will include thinking content blocks, followed by text content blocks.

next_messages = messages + [("ai", ai_msg.content), ("human", "I love AI")]

ai_msg = llm.invoke(next_messages)
ai_msg.content_blocks

[{'type': 'reasoning',
  'reasoning': 'The user wants me to translate "I love AI" from English to French. \n\n"I love" translates to "J\'aime" in French.\n"AI" stands for "Artificial Intelligence" which in French is "Intelligence Artificielle" or "IA" (the French abbreviation).\n\nSo the translation would be "J\'aime l\'IA" or "J\'aime l\'intelligence artificielle".\n\nI think using the abbreviation "IA" would be more natural and concise, similar to how the user used "AI" in English.',
  'extras': {'signature': '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'}},
 {'type': 'text', 'text': "J'aime l'IA."}]

Prompt caching

Bedrock supports caching of elements of your prompts, including messages and tools. This allows you to re-use large documents, instructions, few-shot documents, and other data to reduce latency and costs.

Not all models support prompt caching. See supported models here.

To enable caching on an element of a prompt, mark its associated content block using the cachePoint key. See example below:

import requests
from langchain_aws import ChatBedrockConverse

llm = ChatBedrockConverse(model="us.anthropic.claude-sonnet-4-5-20250929-v1:0")

# Pull LangChain readme
get_response = requests.get(
    "https://raw.githubusercontent.com/langchain-ai/langchain/master/README.md"
)
readme = get_response.text

messages = [
    {
        "role": "user",
        "content": [
            {
                "type": "text",
                "text": "What's LangChain, according to its README?",
            },
            {
                "type": "text",
                "text": f"{readme}",
            },
            {
                "cachePoint": {"type": "default"},
            },
        ],
    },
]

response_1 = llm.invoke(messages)
response_2 = llm.invoke(messages)

usage_1 = response_1.usage_metadata["input_token_details"]
usage_2 = response_2.usage_metadata["input_token_details"]

print(f"First invocation:\n{usage_1}")
print(f"\nSecond:\n{usage_2}")

First invocation:
{'cache_creation': 1528, 'cache_read': 0}

Second:
{'cache_creation': 0, 'cache_read': 1528}

Citations

Citations can be generated if they are enabled on input documents. Documents can be specified in Bedrock’s native format or LangChain’s standard types:

Bedrock format

from langchain_aws import ChatBedrockConverse

llm = ChatBedrockConverse(model="us.anthropic.claude-sonnet-4-20250514-v1:0")

pdf_path = "path/to/your/file.pdf"

with open(pdf_path, "rb") as f:
    pdf_bytes = f.read()

document = {
    "document": {
        "format": "pdf",
        "source": {"bytes": pdf_bytes},
        "name": "my-pdf",
        "citations": {"enabled": True},  
    },
}

response = llm.invoke(
    [
        {
            "role": "user",
            "content": [
                {"type": "text", "text": "Describe this document."},
                document,
            ]
        },
    ]
)
response.content_blocks

---

API reference

For detailed documentation of all ChatBedrock features and configurations head to the API reference: python.langchain.com/api_reference/aws/chat_models/langchain_aws.chat_models.bedrock.ChatBedrock.html

For detailed documentation of all ChatBedrockConverse features and configurations head to the API reference: python.langchain.com/api_reference/aws/chat_models/langchain_aws.chat_models.bedrock_converse.ChatBedrockConverse.html

---

Edit this page on GitHub or file an issue.

Connect these docs to Claude, VSCode, and more via MCP for real-time answers.