Amazon Bedrock customers can now attribute model inference usage to specific teams, applications, environments, and experiments at the individual request level on the InvokeModel and InvokeModelWithResponseStream APIs. This gives customers fine-grained visibility into how their Amazon Bedrock usage is distributed across their organization, helping them
understand consumption patterns, optimize spend, and report usage back to internal stakeholders without provisioning additional resources.

This launch builds on Amazon Bedrock’s existing portfolio of usage attribution capabilities. Customers can already attribute model inference usage at the resource and identity level using application inference profiles, IAM principal-based attribution, project-level tracking on the OpenAI-compatible bedrock-mantle endpoint, and workspace-level tracking for
Anthropic Claude models. For finer-grained, per-request attribution, the Converse and ConverseStream APIs have supported request-level metadata since launch. Today’s release brings the same capability to the InvokeModel and InvokeModelWithResponseStream APIs, giving customers a consistent way to tag inference calls across the entire bedrock-runtime endpoint.

With this launch, customers can tag each Amazon Bedrock model inference call with attributes like team, project, or environment, and analyze usage by these tags in Amazon Bedrock model invocation logs. To get started, enable model invocation logging in the AWS Region where you call Amazon Bedrock, then add metadata to your inference requests. This feature is available in all AWS commercial Regions where Amazon Bedrock is available. To learn more, see Request metadata.