Inference using Anthropic Messages API
The Anthropic Messages API provides native access to Claude models using the Anthropic
request and response format. You can use the Messages API on both the
bedrock-runtime and bedrock-mantle endpoints. For new
applications, we recommend the bedrock-runtime endpoint.
| Endpoint |
Base URL |
Use case |
bedrock-runtime |
https://bedrock-runtime.{region}.amazonaws.com/anthropic |
Recommended for new applications. Uses AWS SDK integrations, IAM authentication, and invocation logging. |
bedrock-mantle |
https://bedrock-mantle.{region}.api.aws/anthropic/v1/messages |
Supports Amazon Bedrock API key authentication and Workspaces (Anthropic-compatible). |
Supported Regions
The Messages API is available on the bedrock-runtime endpoint in all
Regions where Claude models are available. See Regional availability by models for details. On the bedrock-mantle
endpoint, it is available in all Regions that support bedrock-mantle (see
Supported Regions and Endpoints).
Prerequisites
Before using the Messages API, make sure you have the following:
-
Model access – Request access to
Claude models in the Amazon Bedrock console. See Request access to models.
-
Authentication – Depends on the endpoint:
-
bedrock-runtime – Use AWS SigV4 credentials
through the AWS SDK (boto3, etc.), a Amazon Bedrock API key (see API keys), or a short-term Bedrock bearer token generated
from AWS credentials with the aws-bedrock-token-generator library.
-
bedrock-mantle – Use a Amazon Bedrock API key (see API keys) or AWS SigV4 credentials.
-
Anthropic version header – All requests require an API version:
-
bedrock-runtime: Set "anthropic_version": "bedrock-2023-05-31" in the request body when calling InvokeModel. The Anthropic SDK adds the required header automatically when using base_url=https://bedrock-runtime.{region}.amazonaws.com/anthropic.
-
bedrock-mantle: Set anthropic-version: 2023-06-01 as an HTTP header.
Basic request
To send a basic message to a Claude model, choose the tab for your preferred method, and then follow the steps:
- bedrock-runtime (Python, Anthropic SDK)
-
Use the Anthropic SDK against the bedrock-runtime
endpoint's /anthropic route with a short-term Bedrock bearer
token generated by aws-bedrock-token-generator:
from anthropic import Anthropic
from aws_bedrock_token_generator import provide_token
token = provide_token(region="us-east-1")
client = Anthropic(
base_url="https://bedrock-runtime.us-east-1.amazonaws.com/anthropic",
api_key=token,
)
response = client.messages.create(
model="global.anthropic.claude-sonnet-5",
max_tokens=1024,
messages=[
{"role": "user", "content": "Explain quantum computing in one sentence."}
]
)
print(response.content[0].text)
- bedrock-runtime (Python, boto3)
-
import boto3
import json
client = boto3.client("bedrock-runtime", region_name="us-east-1")
response = client.invoke_model(
modelId="anthropic.claude-sonnet-5",
body=json.dumps({
"anthropic_version": "bedrock-2023-05-31",
"max_tokens": 1024,
"messages": [
{"role": "user", "content": "Explain quantum computing in one sentence."}
]
})
)
result = json.loads(response["body"].read())
print(result["content"][0]["text"])
- bedrock-runtime (AWS CLI)
-
Use the AWS CLI to invoke the model:
aws bedrock-runtime invoke-model \
--model-id anthropic.claude-sonnet-5 \
--body '{
"anthropic_version": "bedrock-2023-05-31",
"max_tokens": 1024,
"messages": [
{"role": "user", "content": "Explain quantum computing in one sentence."}
]
}' \
--cli-binary-format raw-in-base64-out \
output.json
- bedrock-runtime (curl)
-
You can call bedrock-runtime directly with curl by signing
the request with AWS SigV4. The example below uses curl's built-in
--aws-sigv4 flag (available in curl 7.75.0 and later):
curl -X POST \
"https://bedrock-runtime.us-east-1.amazonaws.com/model/anthropic.claude-sonnet-5/invoke" \
-H "Content-Type: application/json" \
--aws-sigv4 "aws:amz:us-east-1:bedrock" \
--user "$AWS_ACCESS_KEY_ID:$AWS_SECRET_ACCESS_KEY" \
-d '{
"anthropic_version": "bedrock-2023-05-31",
"max_tokens": 1024,
"messages": [
{"role": "user", "content": "Explain quantum computing in one sentence."}
]
}'
If you are using temporary credentials from AWS STS (for example,
an assumed role), also pass the session token by adding
-H "X-Amz-Security-Token: $AWS_SESSION_TOKEN" to the
request. For most use cases we recommend the AWS SDKs or CLI, which
handle request signing and credential refresh automatically.
- bedrock-runtime (curl, API key)
-
You can also call bedrock-runtime with a Amazon Bedrock API key using
the Anthropic-native /anthropic/v1/messages route, just like
bedrock-mantle:
curl -X POST https://bedrock-runtime.us-east-1.amazonaws.com/anthropic/v1/messages \
-H "x-api-key: $AWS_BEARER_TOKEN_BEDROCK" \
-H "anthropic-version: 2023-06-01" \
-H "Content-Type: application/json" \
-d '{
"model": "us.anthropic.claude-sonnet-5",
"max_tokens": 1024,
"messages": [
{"role": "user", "content": "Explain quantum computing in one sentence."}
]
}'
- bedrock-mantle (curl)
-
curl -X POST https://bedrock-mantle.us-east-1.api.aws/anthropic/v1/messages \
-H "x-api-key: $BEDROCK_API_KEY" \
-H "anthropic-version: 2023-06-01" \
-H "Content-Type: application/json" \
-d '{
"model": "anthropic.claude-sonnet-5",
"max_tokens": 1024,
"messages": [
{"role": "user", "content": "Explain quantum computing in one sentence."}
]
}'
Stream responses
To receive response tokens incrementally as they are generated, choose the tab for your preferred method, and then follow the steps:
- bedrock-runtime (Python, Anthropic SDK)
-
from anthropic import Anthropic
from aws_bedrock_token_generator import provide_token
token = provide_token(region="us-east-1")
client = Anthropic(
base_url="https://bedrock-runtime.us-east-1.amazonaws.com/anthropic",
api_key=token,
)
with client.messages.stream(
model="global.anthropic.claude-sonnet-5",
max_tokens=1024,
messages=[
{"role": "user", "content": "Write a short poem about cloud computing."}
],
) as stream:
for text in stream.text_stream:
print(text, end="", flush=True)
- bedrock-runtime (Python, boto3)
-
import boto3
import json
client = boto3.client("bedrock-runtime", region_name="us-east-1")
response = client.invoke_model_with_response_stream(
modelId="anthropic.claude-sonnet-5",
body=json.dumps({
"anthropic_version": "bedrock-2023-05-31",
"max_tokens": 1024,
"messages": [
{"role": "user", "content": "Write a short poem about cloud computing."}
]
})
)
for event in response["body"]:
chunk = json.loads(event["chunk"]["bytes"])
if chunk["type"] == "content_block_delta":
print(chunk["delta"]["text"], end="")
- bedrock-runtime (curl, API key)
-
curl -X POST https://bedrock-runtime.us-east-1.amazonaws.com/anthropic/v1/messages \
-H "x-api-key: $AWS_BEARER_TOKEN_BEDROCK" \
-H "anthropic-version: 2023-06-01" \
-H "Content-Type: application/json" \
-d '{
"model": "us.anthropic.claude-sonnet-5",
"max_tokens": 1024,
"stream": true,
"messages": [
{"role": "user", "content": "Write a short poem about cloud computing."}
]
}'
- bedrock-mantle (curl)
-
curl -X POST https://bedrock-mantle.us-east-1.api.aws/anthropic/v1/messages \
-H "x-api-key: $BEDROCK_API_KEY" \
-H "anthropic-version: 2023-06-01" \
-H "Content-Type: application/json" \
-d '{
"model": "anthropic.claude-sonnet-5",
"max_tokens": 1024,
"stream": true,
"messages": [
{"role": "user", "content": "Write a short poem about cloud computing."}
]
}'
Supported features
In addition to the Extended
thinking, Adaptive
thinking, Structured
outputs, Compaction, Fallback credit for refused requests (beta), and
Mid-conversation system
messages sections above, the Messages API supports the following features with
Claude models:
-
System prompts – Set model behavior with a system parameter.
-
Multi-turn conversations – Pass alternating user and assistant messages.
-
Tool use – Define tools the model can call. See Use a tool to complete an Amazon Bedrock model response.
-
Vision – Send images in the content array alongside text.
-
Prompt caching – Cache frequently used context to reduce latency and cost.
For the full Messages API request and response field reference, see Request and Response.
Count tokens
To count the number of input tokens that a request would consume before sending it for inference, use the Anthropic count_tokens path on the bedrock-mantle endpoint. This is the only token-counting path supported for Claude models that are not available on bedrock-runtime with a Region-specific endpoint, including Claude models that launch with cross-Region inference (CRIS) only. For details and an example, see Count tokens using the bedrock-mantle endpoint.
