How to use Claude, Claude Instant, Claude 2.x, and the Claude 3 series (Haiku, Sonnet, Opus) with AWS Bedrock

You’re looking to leverage the powerful Anthropic Claude models within AWS Bedrock – that’s a great choice for various generative AI tasks! Using Claude with Bedrock simplifies access and integration compared to direct API calls to Anthropic.

Here’s a breakdown of how to use Claude models on AWS Bedrock:

1. Accessing Claude Models via AWS Bedrock

Before you can use Claude, you need to ensure it’s enabled in your AWS account.

Enable Model Access:
- Go to the AWS Bedrock console.
- In the left-hand navigation pane, find “Model access” under “Getting started” or “Playgrounds.”
- Locate the Anthropic Claude models (e.g., Claude Instant, Claude 2, Claude 3 Haiku, Sonnet, Opus) and request access. This might take a few minutes to process.

2. Interacting with Claude Models

Once access is granted, you have several ways to interact with Claude through Bedrock:

a. Bedrock Playgrounds (for quick testing and exploration)

The Bedrock console offers interactive playgrounds to test models without writing code.

Text Playground:
1. Navigate to “Playgrounds” > “Text” in the Bedrock console.
2. Select “Anthropic” as the category and choose your desired Claude model (e.g., Claude 3 Sonnet).
3. Enter your prompt in the text area. Claude models generally prefer a conversational or instruction-based prompt format.
  - Example Prompt:Human: Write a short, engaging blog post about the benefits of remote work. Assistant:
4. Configure inference parameters on the right side:
  - Temperature: Controls randomness (higher = more creative, lower = more focused).
  - Top P: Controls diversity (higher = more diverse token selection).
  - Max tokens to generate: Sets the maximum length of the response.
  - Stop sequences: Define sequences (e.g., \n\nHuman:) where the model should stop generating text.
5. Click “Run” to generate the response.
Chat Playground:
1. Navigate to “Playgrounds” > “Chat” in the Bedrock console.
2. Select “Anthropic” and your Claude model.
3. This playground is designed for multi-turn conversations, maintaining context. Type your messages in the chat interface.

b. AWS SDKs (for programmatic integration)

For building applications, you’ll use AWS SDKs (e.g., Boto3 for Python, AWS SDK for JavaScript). This is the most common way to integrate Claude into your own software.

Key API Call: The primary API you’ll use is invoke_model.
Example using Python (Boto3):Pythonimport boto3 import json # Initialize the Bedrock Runtime client bedrock_runtime = boto3.client( service_name='bedrock-runtime', region_name='us-east-1' # Replace with your AWS region ) # --- Define the model and prompt --- model_id = "anthropic.claude-3-sonnet-20240229-v1:0" # Or 'anthropic.claude-3-haiku-20240307-v1:0', 'anthropic.claude-2-1', etc. # Claude 3 models use the 'messages' API format (recommended) # Claude 2.x and Instant use the 'text_prompt' format (legacy, but still supported) # Example for Claude 3 (Haiku, Sonnet, Opus) body_claude_3 = json.dumps({ "messages": [ {"role": "user", "content": "Tell me a fun fact about giraffes."}, ], "anthropic_version": "bedrock-2023-05-31", # Required for Claude 3 "max_tokens": 300, "temperature": 0.7, "top_p": 0.9 }) # Example for Claude 2.x and Instant (legacy text_prompt format) # This format is still supported for Claude 3 but 'messages' is preferred for new development. body_claude_2 = json.dumps({ "prompt": "Human: Tell me a fun fact about giraffes.\n\nAssistant:", "max_tokens_to_sample": 300, "temperature": 0.7, "top_p": 0.9, "stop_sequences": ["\n\nHuman:"] }) try: # --- Invoke the model --- # Choose the appropriate body based on the model_id if "claude-3" in model_id: response = bedrock_runtime.invoke_model( body=body_claude_3, modelId=model_id, accept="application/json", contentType="application/json" ) response_body = json.loads(response.get('body').read()) # Parse Claude 3 response generated_text = response_body['content'][0]['text'] else: response = bedrock_runtime.invoke_model( body=body_claude_2, modelId=model_id, accept="application/json", contentType="application/json" ) response_body = json.loads(response.get('body').read()) # Parse Claude 2.x/Instant response generated_text = response_body['completion'] print(f"Generated Text:\n{generated_text}") except Exception as e: print(f"Error invoking model: {e}")

c. AWS CLI (for command-line users)

You can also interact with Bedrock using the AWS Command Line Interface (CLI).

Bash

# Example for Claude 3 (Haiku, Sonnet, Opus)
aws bedrock-runtime invoke-model \
    --model-id anthropic.claude-3-sonnet-20240229-v1:0 \
    --region us-east-1 \
    --body '{
        "messages": [
            {"role": "user", "content": "Tell me a fun fact about giraffes."}
        ],
        "anthropic_version": "bedrock-2023-05-31",
        "max_tokens": 300,
        "temperature": 0.7
    }' \
    output.json

# View the output (for Claude 3, look inside the 'content' array)
cat output.json

# Example for Claude 2.x and Instant (legacy text_prompt format)
aws bedrock-runtime invoke-model \
    --model-id anthropic.claude-2-1 \
    --region us-east-1 \
    --body '{
        "prompt": "Human: Tell me a fun fact about giraffes.\n\nAssistant:",
        "max_tokens_to_sample": 300,
        "temperature": 0.7,
        "stop_sequences": ["\n\nHuman:"]
    }' \
    output.json

# View the output (for Claude 2.x/Instant, look for 'completion')
cat output.json

3. Key Considerations for Claude Models

Prompt Engineering: Claude models, especially the Claude 3 series, respond exceptionally well to clear, detailed instructions.
- Role-playing: Ask Claude to “Act as a marketing expert…”
- Constraints: “Ensure the response is no more than 100 words.”
- Few-shot examples: Provide examples of desired input/output pairs.
- Chain-of-thought: Prompt the model to think step-by-step.
API Format:
- Claude 3 series (Haiku, Sonnet, Opus): Prefer the messages API format. This is a more structured, chat-optimized format where you provide an array of objects representing “user” and “assistant” turns. It also requires anthropic_version.
- Claude 2.x and Claude Instant: Use the prompt API format (the Human: / Assistant: format). While Bedrock supports the messages format for these models too, the prompt format is what they were originally designed for.
Context Window: Claude models have very large context windows (especially Claude 2.1 and Claude 3 Opus/Sonnet), allowing them to process and generate very long texts. Be mindful of token limits for both input and output.
Cost: Different Claude models have different pricing tiers. Generally, more powerful models (like Claude 3 Opus) are more expensive per token than less powerful ones (like Claude Instant or Haiku).
Fine-tuning: For highly specific tasks, you can fine-tune Claude models on your own datasets within Bedrock to improve performance on your unique use cases.

By following these steps, you can effectively integrate and utilize the powerful capabilities of Anthropic’s Claude models within your AWS-powered applications.