Amazon Bedrock now supports reasoning summaries for OpenAI models

Amazon Bedrock has introduced support for the reasoning.summary parameter for OpenAI models through its Responses API. This feature enables developers to request a concise, human-readable summary of the model’s reasoning process, which accompanies the final response. The summaries are designed to clarify how the model arrived at its answer, particularly for complex tasks like coding, analysis, or multi-step problem solving. This additional context helps users evaluate responses more effectively and debug applications by revealing the model’s decision-making pathway.

The reasoning summaries are returned in the summary array of the reasoning output item, separate from the model’s direct answer. This structure ensures that developers can access both the final output and the underlying reasoning without additional processing. The feature is intended to enhance transparency, making it easier to trust and refine model outputs in production environments. Users can integrate these summaries into their workflows to improve explainability and user communication.

This capability is now available for all OpenAI models on Amazon Bedrock across all AWS Regions where OpenAI GPT models are supported. It functions with in-Region inference, geographic (GEO) cross-Region inference, and global cross-Region inference, ensuring broad accessibility. Developers can leverage this feature regardless of their deployment location, provided they are using an OpenAI model on Amazon Bedrock.

To implement reasoning summaries, users can refer to the OpenAI documentation for detailed guidance on the parameter and its usage. Amazon Bedrock also provides documentation for getting started with OpenAI GPT models, including setup and integration steps. These resources help developers quickly adopt the feature and incorporate reasoning summaries into their applications.