I want my Anthropic Claude Sonnet 4 AI to behave as a reasoning model, but I'm not able to apply the required configuration.

Question

I want reasoning behavior for my Claude Sonnet 4 model, and I am using invoke_inline_agent to make calls to the AI.

I referred to the following pages to configure my invoke_inline_agent:
1. https://docs.aws.amazon.com/bedrock/latest/userguide/inline-agent-invoke.html
2. https://docs.aws.amazon.com/bedrock/latest/userguide/kb-test-configure-reasoning.html?utm_source=chatgpt.com
3. https://docs.aws.amazon.com/bedrock/latest/APIReference/API_agent-runtime_RetrieveAndGenerateStream.html

Configuration Snapshot:

```
agent = InlineAgent(
    foundation_model=model_id,
    instruction=agent_instructions,
    agent_name="agent_name",
    action_groups=[
        groups,
        {
            "name": "CodeInterpreter",
            "builtin_tools": {
                "parentActionGroupSignature": "AMAZON.CodeInterpreter"
            },
        },
    ],
    prompt_override_configuration={
        'promptConfigurations': [
            {
                # "foundationModel": model_id,
                "basePromptTemplate": '100000',
                "parserMode": "DEFAULT",
                "promptCreationMode": "DEFAULT",
                "promptState": "ENABLED",
                "promptType": "ORCHESTRATION",

"additionalModelRequestFields": {
                    "reasoning_config": {
                        "type": "enabled",
                        "budget_tokens": 1024
                    }
                },

"inferenceConfiguration": {
                    "maximumLength": 8000,
                    "temperature": 1,
                    "topP": 1
                },
            }
        ]
    }
)

```

I tried using this configuration. Can you please tell me what might be wrong in the above code snippet for enabling reasoning?

We checked the entire orchestration of the agent, but reasoning Content and thoughts are not visible for Claude Sonnet 4. And also tried for Claude Sonnet 3.7 not working for this too.

Ganesh Thiyagarajan · Answer

Change the *additionalModelRequestFields* with using thinking parameter 
```
"additionalModelRequestFields": {
    "thinking": {
        "type": "enabled",
        "budget_tokens": 1024
    }
}
```
Please refer this [blog ](https://community.aws/content/2tWvN7GNtVuBco4fNgLuowHas2c/how-to-use-reasoning-with-claude-3-7-sonnet-on-amazon-bedrock-python-edition?lang=en)for additional implementation details

re:Post Agent · Answer

To enable reasoning for Claude Sonnet 4 in your InlineAgent configuration, there are a few issues with your current setup that need to be addressed:

1. The reasoning configuration should be placed in the `additionalModelRequestFields` parameter, but it needs to be structured correctly. For Claude Sonnet 4, reasoning can be enabled or disabled using a configurable token budget.

2. The `basePromptTemplate` value of '100000' appears incorrect - this should typically contain your prompt template, not a numeric value.

3. Reasoning is only applied during the generation step, not during the orchestration step. Your current configuration is targeting the "ORCHESTRATION" prompt type.

Here's how you should modify your configuration:

```
prompt_override_configuration={
'promptConfigurations': [
{
# "foundationModel": model_id,
"parserMode": "DEFAULT",
"promptCreationMode": "DEFAULT",
"promptState": "ENABLED",
"promptType": "GENERATION",  # Change from ORCHESTRATION to GENERATION

"additionalModelRequestFields": {
"anthropic_version": "bedrock-2023-05-31",
"reasoning_config": {
"enable_reasoning": True,
"reasoning_token_budget": 1024
}
},

"inferenceConfiguration": {
"maximumLength": 8000,
"temperature": 1,
"topP": 1
},
}
]
}
```

Key changes:
- Changed `promptType` from "ORCHESTRATION" to "GENERATION" since reasoning is only enabled during the generation step
- Updated the reasoning configuration format to match what Claude models expect
- Added the required "anthropic_version" parameter

Note that when using reasoning models, there's a five-minute timeout limit, and the models can use up to 8192 tokens to respond to queries (including both output and thinking tokens). Also, enabling reasoning will increase latency, but it will provide more detailed step-by-step thinking for complex tasks.  
**Sources**  
[Enhance model responses with model reasoning - Amazon Bedrock](https://docs.aws.amazon.com/bedrock/latest/userguide/inference-reasoning.html)  
[Configure response generation for reasoning models and considerations - Amazon Bedrock](https://docs.aws.amazon.com/bedrock/latest/userguide/kb-test-configure-reasoning.html)  
[Introducing Claude 4 in Amazon Bedrock, the most powerful models for coding from Anthropic | AWS News Blog](https://aws.amazon.com/blogs/aws/claude-opus-4-anthropics-most-powerful-model-for-coding-is-now-in-amazon-bedrock/)

I want my Anthropic Claude Sonnet 4 AI to behave as a reasoning model, but I'm not able to apply the required configuration.

Add your answer

Relevant content