Amazon Bedrock costs by model and by team
The CostReveal team, , 4 minute read
Amazon Bedrock is often the easiest way into AI on AWS: the models are a console away, billed on the account you already have. It is also one of the harder lines on the bill to read. Bedrock spend is filed under product codes that differ by model vendor, the model is named only in each line's details, and nothing on the line says which team or product made the call. This post covers how to see Bedrock spend by model and by team with CostReveal, from what AWS already records.
What the AWS bill records for Bedrock
Bedrock spend arrives in the Cost and Usage Report, CUR 2.0, with the rest of your AWS spend. If AWS is already connected to CostReveal, there is nothing more to connect. The export CostReveal sets up carries what Bedrock needs: resource IDs, the quantities billed, and the identity of each caller.
Each Bedrock line states what it billed, such as input, output, cache read or cache write tokens, and CostReveal records the quantity with the cost, so spend and tokens can be read side by side.
One service, by model
AWS files Bedrock under its Amazon Bedrock product family, but sells different vendors' models under different product codes. Read as they come, one model family's spend scatters across several services.
So every line in the Bedrock product family is shown as one service, Bedrock, whichever vendor's model it is. Its model is AWS's own name for it, as the Bedrock console's model catalog shows it, such as Claude Sonnet 4.6 or Nova Pro. The model comes from AWS itself, so it carries the DIRECT basis. Explore then groups Bedrock spend by model, by account and by region, as precisely as AWS billed it.
By team: who made the call
A Bedrock call is made by an IAM user or role, and the export splits Bedrock lines by the principal that made each call. That is the key to spend by team.
- Tag the principals in IAM, with keys such as
teamorcustomer. A role a service assumes to call Bedrock is the natural place: the payments service's role taggedteam = payments. - Activate the tags in the AWS Billing console under Cost allocation tags, filtered to IAM principal tags.
- CostReveal reads them with your project's tag keys, as it reads resource tags, and the spend that role made is the payments team's, with the INFERRED basis.
A tag on the resource itself comes first. An application inference profile, or a Bedrock project, tagged for a team or customer explains its own spend; the caller's tag explains what the resource's tags do not. So a team can give each product its own inference profile and tag it, and keep a shared role for everything else.
AWS lists a principal's tags for activation only after that principal has made a call, so tag the role before it goes to production and activate the tag once the first calls appear.
What stays unknown
Bedrock spend that neither a resource tag nor a caller's tag names is UNKNOWN on that dimension, shown as such and never spread across teams. Two common causes:
- One role for everything. If every service calls Bedrock as the same role, its tag names one owner for all of it. Separate roles, or tagged inference profiles, split it.
- Tags not activated. A tag present in IAM but not activated for cost allocation never reaches the bill.
Home's next best action names the one change that would explain the most spend, wherever it is.
Usage data and Bedrock
Usage data, the record of each call your application sends, divides Anthropic and OpenAI spend by customer. It does not divide Bedrock spend. For Bedrock, the split by team or customer comes from the roles and resources you tag, so a customer that needs its own figure gets its own inference profile or its own role.
Bedrock beside the rest of your AI spend
Bedrock rarely runs alone. Connect Anthropic or OpenAI directly as well, and Explore shows all AI spend together, by provider, then by model: Claude on Bedrock beside Claude through Anthropic's API, each billed where it was billed, never counted twice. A budget can watch Bedrock as a service, a team, or a saved view of one model, and alert before its limit is passed.
Frequently asked questions
Do I need to connect anything besides AWS? No. Bedrock spend arrives through the same export.
How does CostReveal know which model a line is for? From AWS's own model name for it, the one the Bedrock console's catalog shows.
How do I see Bedrock spend by team? Tag the IAM roles that call Bedrock, activate the tags for cost allocation, and CostReveal reads them as it reads resource tags. A tagged inference profile comes first.
Does usage data divide Bedrock spend by customer? No. Use separate roles or tagged inference profiles for the customers that need their own figure.
Why does some Bedrock spend show as unknown? No resource tag or caller's tag names its owner. It stays unknown, in the open, until one does.
Keep reading
- Connecting AWS to CostReveal: the export, the read only role, and Bedrock's details.
- How to track Claude API costs per customer and feature: Claude through Anthropic's API.
- Why CostReveal shows the spend it cannot explain: how unknown spend stays in the open.