Amazon Bedrock Adds Reasoning Summaries for OpenAI Models
Amazon Bedrock now supports reasoning.summary for OpenAI models through the Responses API, offering human-readable summaries for debugging and evaluation.
AWS announced on October 9, 2026 that Amazon Bedrock supports the reasoning.summary parameter for OpenAI models through the Responses API.
Developers can request a human-readable summary of a model's reasoning alongside its answer. The summary is returned in the summary array of a reasoning output item.
This feature does not reveal a complete internal chain of thought. A generated summary is an explanatory artifact, not a definitive audit record or proof that an answer is correct.
Summaries may help developers investigate failures in coding, analysis and multi-step tasks. They are most useful when compared with prompts, tool results and independent reference answers.
AWS says the feature is available where its OpenAI GPT models are offered, including supported regional and cross-region inference arrangements. Availability should be checked for the intended model and region.
Regulated organizations may use summaries as one part of an explanation, but consequential decisions still require evidence, applicable rules and human review.
Logging these summaries raises privacy and retention questions. Sensitive content should be protected through access controls and minimization.
The feature is a useful addition to AI observability, provided developers do not mistake plausible explanations for verified reasoning.