AWS maps out how teams can manage Bedrock spending
A cost-management playbook connects spending attribution, request visibility, budget enforcement and regular operational review.
AWS’s playbook gives cloud-cost specialists and platform engineers a shared operating sequence for Bedrock: attribute spending, observe requests, enforce budgets, optimize defaults and review operations. Cost specialists lead attribution and ongoing review; engineers lead the middle stages. AWS places attribution and observation before optimization, so teams understand which workloads generated their spending.
The logging documentation makes request visibility concrete. Invocation logging starts disabled. Before enabling it, teams must configure an S3 or CloudWatch Logs destination in the same account and Region. Supported calls can supply request data, responses and metadata. This facility covers the bedrock-runtime endpoint, including its OpenAI-compatible Responses and Chat Completions APIs. Calls to bedrock-mantle are excluded, even when they use APIs with those same names.
For bedrock-mantle, Projects provide application-level cost attribution across models. Set the project ID when initializing the client; its requests are then attributed to that project. Requests without an ID use the account’s default project. Project tags must be activated for cost allocation, and can take up to 24 hours to appear afterward. They do not retroactively tag earlier costs. The resulting billed amounts are aggregated: the finest breakdown is by usage type per day, rather than by individual request or prompt.