Code beginner · 3 min read

How to call AWS Bedrock API in Python

Q: How to call AWS Bedrock API in Python

Use the boto3 client for bedrock-runtime to call AWS Bedrock API in Python by invoking the converse or invoke_model methods with the appropriate model ID and message format.

Direct answer

Use the boto3 client for bedrock-runtime to call AWS Bedrock API in Python by invoking the converse or invoke_model methods with the appropriate model ID and message format.

Setup

Install

bash

pip install boto3

Env vars

AWS_ACCESS_KEY_IDAWS_SECRET_ACCESS_KEYAWS_DEFAULT_REGION

Imports

python

import boto3
import json
import os

Examples

inUser message: "Hello, who won the 2024 Olympics?"

outResponse text: "The 2024 Summer Olympics were held in Paris, and the United States topped the medal count."

inUser message: "Explain quantum computing in simple terms."

outResponse text: "Quantum computing uses quantum bits to perform complex calculations much faster than classical computers."

inUser message: "Summarize the latest AI trends."

outResponse text: "Recent AI trends include large multimodal models, improved reasoning capabilities, and wider adoption in industry."

Integration steps

Set up AWS credentials in environment variables or AWS config files.
Import boto3 and create a client for 'bedrock-runtime' with the correct AWS region.
Prepare the message payload in the required JSON format for the model.
Call the 'converse' or 'invoke_model' method with the model ID and message body.
Parse the JSON response to extract the generated text from the model.
Use or display the extracted text as needed in your application.

Full code

python

import boto3
import json
import os

# Initialize the Bedrock client
client = boto3.client('bedrock-runtime', region_name=os.environ.get('AWS_DEFAULT_REGION', 'us-east-1'))

# Define the model ID (example: Anthropic Claude 3.5 Sonnet)
model_id = 'anthropic.claude-3-5-sonnet-20241022-v2:0'

# Prepare the messages payload
messages = [
    {"role": "user", "content": [{"type": "text", "text": "Hello, who won the 2024 Olympics?"}]}
]

# Call the converse API
response = client.converse(
    modelId=model_id,
    messages=messages
)

# Extract the text from the response
output_message = response.get('output', {}).get('message', {})
if output_message and 'content' in output_message and len(output_message['content']) > 0:
    text = output_message['content'][0].get('text', '')
else:
    text = "No response received."

print("Model response:", text)

output

Model response: The 2024 Summer Olympics were held in Paris, and the United States topped the medal count.

API trace

Request

json

{"modelId": "anthropic.claude-3-5-sonnet-20241022-v2:0", "messages": [{"role": "user", "content": [{"type": "text", "text": "Hello, who won the 2024 Olympics?"}]}]}

Response

json

{"output": {"message": {"content": [{"type": "text", "text": "The 2024 Summer Olympics were held in Paris, and the United States topped the medal count."}]}}}

Extractresponse['output']['message']['content'][0]['text']

Variants

Streaming response with AWS Bedrock ›

Use when you want to handle responses incrementally if AWS Bedrock adds streaming support in the future.

python

import boto3
import os

client = boto3.client('bedrock-runtime', region_name=os.environ.get('AWS_DEFAULT_REGION', 'us-east-1'))

model_id = 'anthropic.claude-3-5-sonnet-20241022-v2:0'
messages = [{"role": "user", "content": [{"type": "text", "text": "Explain quantum computing."}]}]

# Note: As of now, AWS Bedrock does not support streaming responses directly via boto3.
# This is a placeholder for future streaming support.

response = client.converse(modelId=model_id, messages=messages)
text = response['output']['message']['content'][0]['text']
print("Streaming (simulated) response:", text)

Async call using aiobotocore ›

Use for asynchronous applications where non-blocking calls to AWS Bedrock improve throughput.

python

import asyncio
import aiobotocore
import os

async def async_bedrock_call():
    session = aiobotocore.get_session()
    async with session.create_client('bedrock-runtime', region_name=os.environ.get('AWS_DEFAULT_REGION', 'us-east-1')) as client:
        model_id = 'anthropic.claude-3-5-sonnet-20241022-v2:0'
        messages = [{"role": "user", "content": [{"type": "text", "text": "Summarize the latest AI trends."}]}]
        response = await client.converse(modelId=model_id, messages=messages)
        text = response['output']['message']['content'][0]['text']
        print("Async response:", text)

asyncio.run(async_bedrock_call())

Invoke alternative model (Llama 3.1) ›

Use when you want to leverage a different model available on AWS Bedrock for varied capabilities or cost.

python

import boto3
import os

client = boto3.client('bedrock-runtime', region_name=os.environ.get('AWS_DEFAULT_REGION', 'us-east-1'))

model_id = 'meta.llama3-1-70b-instruct-v1:0'
messages = [{"role": "user", "content": [{"type": "text", "text": "Write a poem about spring."}]}]

response = client.converse(modelId=model_id, messages=messages)
text = response['output']['message']['content'][0]['text']
print("Llama model response:", text)

Performance

Latency~1-2 seconds per request depending on model and region

Cost~$0.003 to $0.01 per 1,000 tokens depending on model

Rate limitsDefault AWS Bedrock limits vary by account; typically 20-50 requests per second per model

Limit max_tokens to control cost and latency.
Use concise prompts to reduce token usage.
Cache frequent queries to avoid repeated calls.

Approach	Latency	Cost/call	Best for
Synchronous boto3 call	~1-2s	~$0.005	Simple scripts and batch jobs
Async aiobotocore call	~1-2s (non-blocking)	~$0.005	High concurrency apps
Alternative model invocation	~1-3s	~$0.003-$0.01	Use different model capabilities or cost optimization

✓

Quick tip

Always format your messages as a list of role-content dicts with text typed content to comply with AWS Bedrock's expected input schema.

⚠

Common mistake

Passing messages as plain strings or missing the nested content array with type and text fields causes API errors.

Verified 2026-04 · anthropic.claude-3-5-sonnet-20241022-v2:0, meta.llama3-1-70b-instruct-v1:0

Verify ↗

Community Notes

No notes yetBe the first to share a version-specific fix or tip.