Amazon Bedrock expands support for request-level usage attribution

Amazon Web Services has expanded usage attribution capabilities for Amazon Bedrock, allowing customers to track AI model inference usage at the individual request level through the InvokeModel and InvokeModelWithResponseStream APIs. The new feature enables organizations to tag each model inference call with specific metadata such as team, project, application, or environment identifiers, providing granular visibility into how Bedrock resources are consumed across different organizational units. The enhancement builds upon existing attribution methods in Bedrock, including application inference profiles and IAM principal-based tracking. Previously, request-level metadata was only available through the Converse and ConverseStream APIs, but this release extends the capability to the broader bedrock-runtime endpoint for consistent tagging across all inference calls. Organizations can analyze usage patterns through Amazon Bedrock model invocation logs after enabling logging in their AWS region and adding metadata to inference requests. The feature is immediately available across all AWS commercial regions where Amazon Bedrock operates, requiring no additional resource provisioning. This capability addresses enterprise needs for cost allocation, usage optimization, and internal chargeback reporting as AI workloads scale across large organizations with multiple teams and projects utilizing foundation models.