Select your cookie preferences

We use essential cookies and similar tools that are necessary to provide our site and services. We use performance cookies to collect anonymous statistics, so we can understand how customers use our site and make improvements. Essential cookies cannot be deactivated, but you can choose “Customize” or “Decline” to decline performance cookies.

If you agree, AWS and approved third parties will also use cookies to provide useful site features, remember your preferences, and display relevant content, including relevant advertising. To accept or decline all non-essential cookies, choose “Accept” or “Decline.” To make more detailed choices, choose “Customize.”

AWS Logo
Menu
How to use Reasoning with Claude 3.7 Sonnet on Amazon Bedrock - JavaScript Edition

How to use Reasoning with Claude 3.7 Sonnet on Amazon Bedrock - JavaScript Edition

Learn how to use Claude 3.7 Sonnet's step-by-step thinking process with the AWS SDK for JavaScript v3. This tutorial walks you through a practical example showing how to receive and process Claude's internal thought process in your own applications.

Dennis Traub
Amazon Employee
Published Feb 25, 2025
Last Modified Mar 5, 2025
Also available in: C# | Java | Python
Have you ever wondered what goes on inside Claude when answering a complex question? With Claude 3.7 Sonnet's new reasoning feature, you can now peek behind the curtain and see the AI's step-by-step thought process. This capability offers unprecedented transparency into how Claude approaches problems, breaks them down, and reaches conclusions.
In this guide, we'll build a JavaScript application that taps into this powerful capability through Amazon Bedrock. You'll learn how to enable reasoning, allocate appropriate computational resources for it, and process the structured responses that include both Claude's internal reasoning and its final answer. Whether you're debugging prompts, verifying calculations, or using Claude as an educational tool, understanding its reasoning process can significantly enhance your AI interactions.
👨‍💻 Want to jump straight to the code? If you're familiar with Claude and Amazon Bedrock and just need the technical details, skip directly to the step-by-step implementation guide.

What is Reasoning?

Reasoning is Claude's ability to explicitly work through complex problems step by step before providing an answer. When enabled, Claude doesn't just give you the final response – it reveals its complete thought process, similar to how a human might think aloud when solving a problem.
This feature allows Claude to:
  • Break down complex questions into manageable components
  • Explore multiple approaches to solving a problem
  • Analyze information systematically and draw logical conclusions
  • Show its work, particularly valuable for mathematical or logical problems
  • Identify potential flaws in its own thinking
Reasoning is particularly useful for problems that require multi-step thinking, such as complex mathematical calculations, logical puzzles, or nuanced analytical questions. By exposing Claude's internal thought process, you gain transparency into how the AI arrives at its conclusions, which can help build trust and provide educational value.
When reasoning is enabled, Claude allocates a portion of its computational resources (specified by your token budget) to this explicit thinking process, which is then returned to you alongside the final answer. This gives you unprecedented visibility into the AI's problem-solving approach.

Prerequisites

Before we begin, make sure you have:
¹ At the time of writing, Claude 3.7 is available in us-east-1, us-east-2, and us-west-2.

Setting Up Dependencies

You'll need to install the AWS SDK for JavaScript v3. Here's how to include the required dependencies:
1
npm install @aws-sdk/client-bedrock-runtime
Or add the following to your package.json file:
1
2
3
4
5
{
"dependencies": {
"@aws-sdk/client-bedrock-runtime": "^3.755.0"
}
}

Step 1: Create a Basic Project Structure

Create a new JavaScript file (e.g., claude_reasoning.js) where we'll implement our code to interact with Claude's reasoning capability:
1
2
3
import { BedrockRuntimeClient, ConverseCommand } from "@aws-sdk/client-bedrock-runtime";

// We'll implement our code here

Step 2: Set Up the Amazon Bedrock Client

First, we need to create a Bedrock runtime client that will handle our API requests:
1
2
3
4
5
// Create the Amazon Bedrock runtime client
const client = new BedrockRuntimeClient({ region: "us-east-1" });

// Specify the model ID for Claude 3.7 Sonnet
const modelId = "us.anthropic.claude-3-7-sonnet-20250219-v1:0";

Step 3: Create a Message for Claude

Next, we'll create a simple message to send to Claude:
1
2
3
4
5
6
7
8
// Create the message with the user's prompt
const userMessage = "Describe the purpose of a 'hello world' program in one line.";
const conversation = [
{
role: "user",
content: [{ text: userMessage }],
},
];

Step 4: Configure Reasoning Parameters

Here's where the magic happens - we need to enable reasoning and set a token budget for it:
1
2
3
4
5
6
7
// Configure reasoning parameters with a 2000 token budget
const reasoningConfig = {
thinking: {
type: "enabled",
budget_tokens: 2000
}
};
The budget_tokens parameter defines the maximum number of tokens Claude can use for its reasoning process. Adjust this value based on the complexity of your prompt.

Step 5: Send the Request to Claude

Now we'll send our prompt along with the reasoning configuration:
1
2
3
4
5
6
// Send message and reasoning configuration to the model
const response = await client.send(new ConverseCommand({
additionalModelRequestFields: reasoningConfig,
messages: conversation,
modelId
}));
The additionalModelRequestFields parameter is what allows us to pass the reasoning configuration. Notice that we're using the ConverseCommand class from the AWS SDK for JavaScript v3.

Step 6: Process the Response

Claude will return both its reasoning and final response, which we need to extract:
1
2
3
4
5
6
7
8
9
10
11
12
13
14
// Extract the list of content blocks from the model's response
const contentBlocks = response.output.message.content;
let reasoning = null;
let text = null;

// Process each content block to find reasoning and response text
for (const block of contentBlocks) {
if ("reasoningContent" in block) {
reasoning = block.reasoningContent.reasoningText.text;
}
if ("text" in block) {
text = block.text;
}
}
The response contains multiple content blocks, with the reasoning in one block and the final response text in another.

Step 7: Display the Results

Finally, we'll display both the reasoning process and the final answer:
1
2
3
4
5
6
7
8
if (reasoning) {
console.log("<thinking>");
console.log(reasoning);
console.log("</thinking>\n");
}
if (text) {
console.log(text);
}

Complete Example

Here's the complete working example
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
import { BedrockRuntimeClient, ConverseCommand, } from "@aws-sdk/client-bedrock-runtime";

/**
* This example demonstrates how to use Anthropic Claude 3.7 Sonnet's reasoning capability
* with Amazon Bedrock. It shows how to:
* - Set up the Amazon Bedrock runtime client
* - Create a message
* - Configure reasoning parameters
* - Send a request with reasoning enabled
* - Process both the reasoning output and final response
*/


// Create the Amazon Bedrock runtime client
const client = new BedrockRuntimeClient({ region: "us-east-1" });

// Specify the model ID. For the latest available models, see:
// https://docs.aws.amazon.com/bedrock/latest/userguide/models-supported.html
const modelId = "us.anthropic.claude-3-7-sonnet-20250219-v1:0";

// Create the message with the user's prompt
const userMessage = "Describe the purpose of a 'hello world' program in one line.";
const conversation = [
{
role: "user",
content: [{ text: userMessage }],
},
];

// Configure reasoning parameters with a 2000 token budget
const reasoningConfig = {
thinking: {
type: "enabled",
budget_tokens: 2000
}
};

try {
// Send message and reasoning configuration to the model
const response = await client.send(new ConverseCommand({
additionalModelRequestFields: reasoningConfig,
messages: conversation,
modelId
}));

// Extract both reasoning and final response
const contentBlocks = response.output.message.content;
let reasoning = null;
let text = null;

// Process each content block to find reasoning and response text
for (const block of contentBlocks) {
if ("reasoningContent" in block) {
console.log("<thinking>");
console.log(block.reasoningContent.reasoningText.text);
console.log("</thinking>\n");
}
if ("text" in block) {
console.log(block.text);
}
}

} catch (err) {
console.log(`ERROR: Can't invoke '${modelId}'. Reason: ${err}`);
process.exit(1);
}
Note: Make sure to run this code with a recent version of Node.js that supports ES modules. You may need to add "type": "module" to your package.json file or use the .mjs extension for your JavaScript file.

Example Output

When you run this code, you'll see output similar to:
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
<thinking>
I need to explain the purpose of a "Hello World" program in a single line.

A "Hello World" program is typically the simplest program that can be written in
a programming language. Its purpose is to:
1. Verify that the programming environment is set up correctly
2. Demonstrate the minimal syntax needed to create a functioning program
3. Provide a simple introduction to the language
4. Output the text "Hello, World!" or a similar message

I need to condense this information into a concise, one-line description that
captures the essence of what a "Hello World" program is for.
</thinking>

A "Hello World" program serves as the simplest introduction to a programming
language by demonstrating basic syntax while verifying the environment works correctly.

Summary and next steps

The reasoning capability transforms how we interact with AI by providing valuable insight into the model's thinking process. By following the steps in this guide, you can now access and leverage this powerful feature through Amazon Bedrock using JavaScript.
The reasoning feature opens up exciting possibilities to enhance trust, improve outcomes, and gain deeper insights from your AI interactions. For more advanced use cases, consider:
  • Adjusting the token budget based on your problem complexity – more complex problems may benefit from larger reasoning budgets
  • Using the reasoning output to validate multi-step calculations or complex analytical processes
  • Comparing different reasoning approaches by adjusting your prompts
  • Integrating reasoning with other Claude capabilities like function calling for powerful, transparent AI solutions
  • Using reasoning as an educational tool to understand expert-level problem-solving approaches
By incorporating reasoning into your applications, you're not just getting answers – you're gaining insight into the full problem-solving journey. This transparency can help build trust with users and provide richer, more valuable AI interactions.
Happy coding!
 

Any opinions in this post are those of the individual author and may not reflect the opinions of AWS.

Comments

Log in to comment