Extended thinking in Amazon Nova 2

Amazon Nova 2 Lite introduces extended thinking capabilities that enable the model to engage in deeper reasoning for complex problems. This optional feature gives you control over when and how the model allocates additional computational resources to think through challenging tasks.

How Extended Thinking Works

Amazon Nova 2 introduces extended thinking as a hybrid capability. You have complete control:

Reasoning effort

The following shows different levels of reasoning effort.

Low Effort (maxReasoningeffort:"low")

Best for: Tasks with added complexity requiring structured thinking. For example, you can use this for code review and improvement suggestions where the model needs to carefully consider existing code quality, analysis tasks that require thoughtful consideration of multiple factors, or problem-solving scenarios that benefit from a methodical approach. Low effort is ideal for compound tasks where basic reasoning improves accuracy without requiring deep multi-step planning.

Medium effort (maxReasoningEffort: "medium")

Best for: Multi-step tasks and coding workflows. For example, you can use this for software development and debugging where the model needs to understand existing code structure before implementing changes, code generation that requires coordination across multiple files or components, multi-step calculations with interdependencies, or planning tasks with multiple constraints. Medium effort is optimal for agentic workflows that coordinate multiple tools and require the model to maintain context across several sequential operations.

High Effort (maxReasoningeffort:"high")

Best for: STEM reasoning and advanced problem-solving. For example, you can use this for advanced mathematical problems and proofs that require rigorous step-by-step verification, scientific analysis and research tasks demanding deep investigation, complex system design with architectural considerations across multiple dimensions, or critical decision-making scenarios with significant implications. High effort delivers maximum accuracy for tasks requiring sophisticated reasoning, careful evaluation of alternatives, and thorough validation of conclusions.

Quick start: Enabling Extended Thinking

Extended thinking is controlled through the reasoningConfig parameter.

import boto3

bedrock = boto3.client('bedrock-runtime', region_name='us-east-1')

response = bedrock.converse(
    modelId='us.amazon.nova-2-lite-v1:0',
    system=[{"text": "You are a highly capable personal assistant"}],
    messages=[{
        "role": "user",
        "content": [{"text": "Provide a meal plan for a gluten free family of 4."}]
    }],
    inferenceConfig={
        "temperature": 0.7,
        "topP": 0.9,
        "maxTokens": 10000
    },
    additionalModelRequestFields={
        "reasoningConfig": {
            "type": "enabled",
            "maxReasoningEffort": "low"
        }
    }
)

content_list = response["output"]["message"]["content"]

for item in content_list:
    if "reasoningContent" in item:
        reasoning_text = item["reasoningContent"]["reasoningText"]["text"]
        print("=== REASONING ===")
        print(reasoning_text)
        print()
    elif "text" in item:
        print("=== ANSWER ===")
        print(item["text"])

Reasoning Parameters:

For full examples of code that utilizes extended thinking, see Code library.

Response Structure

When you enable extended thinking, responses include reasoningContent blocks followed by text content blocks:

{
    "output": {
        "message": {
            "role": "assistant",
            "content": [
                {
                    "reasoningContent": {
                        "reasoningText": {
                            "text": "[REDACTED]"
                        }
                    }
                },
                {
                    "text": "Based on the premises, we can conclude..."
                }
            ]
        }
    },
    "stopReason": "end_turn"
}

Configuration Options

Amazon Nova 2 introduces a new reasoningConfig parameter that you can add to your existing converse request structure to enable reasoning:

additionalModelRequestFields={
    "reasoningConfig": {
        "type": "enabled",  # or "disabled" (default)
        "maxReasoningEffort": "high"  # "low", "medium", or "high"
    }
}

Parameters:

Supported models

Extended thinking is currently available in: Amazon Nova 2 Lite (us.amazon.nova-2-lite-v1:0)

Extended thinking with tool use

Extended thinking works seamlessly with tool calling, allowing Amazon Nova to reason about which tools to use and how to interpret their results.

Understanding Reasoning Tokens and Pricing

Token types

Extended thinking tokens are billed as output tokens:

Usage breakdown

All three token types are included in your usage metrics and billing. Reasoning tokens are priced the same as output tokens and will appear as "REDACTED" in the model response".

  {
  "usage": {
    "inputTokens": 45,
    "outputTokens": 1240,
    "totalTokens": 1285
  }
}
  

Frequently Asked Questions