---
title: How do you track Claude spend across Amazon Bedrock, Claude Platform on AWS and the Anthropic API?
url: https://opencontroller.lyzr.com/blog/track-claude-spend-bedrock-vs-anthropic-api
published: 2026-11-11
updated: 2026-11-11
author: Lyzr
---

# How do you track Claude spend across Amazon Bedrock, Claude Platform on AWS and the Anthropic API?

**Each channel reports Claude spend in a different place. Anthropic API spend comes from the Usage and Cost Admin API, which needs an Admin API key; Amazon Bedrock spend sits in AWS billing as a native AWS service; Claude Platform on AWS is billed through AWS Marketplace and broken down in the Claude Console. None of them knows which agent made a call, so you attribute spend by giving each agent its own key, workspace or inference profile, or by routing traffic through a gateway.**

Large organisations rarely buy Claude one way. A platform team uses Bedrock, a product team has a direct Anthropic account, and someone has subscribed to Claude Platform on AWS. Finance then asks for one number. This guide shows where each number lives.

## Where does Claude spend show up in each channel?

| Channel | Who runs inference | Where cost appears | Access you need | Breakdown available |
|---|---|---|---|---|
| Anthropic API (Claude Console) | Anthropic | Usage and Cost Admin API; Console Usage and Cost pages | Admin API key (prefix `sk-ant-admin01-`) or other admin credential | Usage by API key, workspace, model, service tier, inference geography; cost by workspace or description |
| Claude Platform on AWS | Anthropic | AWS bill via AWS Marketplace, as Claude Consumption Units; Claude Console Usage and Cost pages | AWS IAM permissions plus Console access | Console shows model and workspace; Cost Explorer shows one aggregated line item |
| Amazon Bedrock | AWS | AWS billing as an AWS service; Cost Explorer; CloudWatch metrics | IAM read access to billing and Bedrock | Account, region, model; finer split via application inference profiles and tags |
| Google Vertex AI | Google Cloud | Google Cloud Billing, including billing export to BigQuery | Billing viewer on the billing account | Project, SKU, labels |

## What do you need to know about the Anthropic Usage and Cost API?

- **Endpoints.** Usage is `GET /v1/organizations/usage_report/messages`; cost is `GET /v1/organizations/cost_report`. Anthropic's docs now describe these as legacy but supported, and point new integrations to its Usage and Cost Analytics API.
- **Credential.** Workspace API keys do not work. You need an Admin API key or an equivalent admin credential.
- **Granularity.** Usage buckets can be 1 minute, 1 hour or 1 day. Cost is daily only, in US dollars.
- **Freshness.** Data typically appears within about five minutes of a request, and Anthropic asks callers to poll no more than once a minute for sustained use.
- **Gaps.** Priority Tier costs are excluded from the cost endpoint, so track them through usage grouped by service tier. The API is not available to claude.ai Team plans or individual Console accounts, and Claude Enterprise parent organisations use an Analytics API key instead.

If you cannot obtain an admin key, see our guide to [discovering AI usage without admin API keys](/blog/discover-ai-usage-without-admin-api-keys).

## How is Claude Platform on AWS different from Amazon Bedrock for cost tracking?

Claude Platform on AWS gives you Anthropic's own platform through your AWS account. Anthropic runs inference; AWS handles authentication, IAM and billing.

- **Billing unit.** Usage is metered hourly in Claude Consumption Units, billed through AWS Marketplace and invoiced monthly in arrears. CCUs are not prepaid credits.
- **Visibility.** AWS Cost Explorer shows the aggregated CCU line. The detailed model and workspace view is in the Claude Console Usage and Cost pages.
- **Lag.** Spend can take about two hours to appear, so usage can pass a spend cap before requests fail.
- **API.** The Usage and Cost API endpoints are not available on Claude Platform on AWS.
- **Audit.** CloudTrail records management events by default. Inference calls are data events, which you must enable explicitly and which cost extra.

On Amazon Bedrock, AWS runs inference and Claude is billed as part of Bedrock. Bedrock publishes token and invocation metrics to CloudWatch, model invocation logging can write request records to CloudWatch Logs or S3 once you turn it on, and application inference profiles let you tag model usage so Cost Explorer can split it by team or application.

## How do you combine the channels into one view?

1. **Find every channel.** Check each AWS account for Bedrock usage and Marketplace subscriptions, each Anthropic organisation and workspace, and each Google Cloud project with Vertex AI enabled.
2. **Get read-only access per channel.** Admin key for Anthropic, IAM billing read for AWS, billing viewer for Google Cloud.
3. **Normalise to one schema.** Date, channel, account or workspace, model, input tokens, output tokens, cost.
4. **Map to agents.** Native sources stop at key, workspace, profile or project. Give each production agent its own key, workspace or inference profile, or route it through a gateway that tags every call with the agent and version.
5. **Reconcile monthly** against the invoices, so the dashboard and the bill agree.

Step 4 is where most programmes stall. Our guide to [attributing LLM costs to AI agents](/blog/how-to-attribute-llm-costs-to-ai-agents) covers the options, and [LLM gateway vs agent gateway vs control plane](/blog/llm-gateway-vs-agent-gateway-vs-control-plane) explains where a gateway fits.

## How Opencontroller does this

Opencontroller auto-discovers agents, models and API keys across AWS, Azure and Google Cloud, and connects to Claude Enterprise workspaces, so Claude usage bought through different channels lands in one inventory. Its LLM Gateway (Lyzr Gateway) exposes OpenAI-compatible endpoints with routing, fallback and retry, caching, and cost attribution per agent and per version, which fills the agent-level gap that native billing leaves. In Lyzr's October 2026 benchmark, caching gave 23.5% lower model cost on mixed synthetic traffic, under the stated test conditions. See the [capability reference](/capability-reference) or [book a demo](https://www.lyzr.ai/book-demo/).

## Frequently asked questions

**Which key do I need to read Anthropic API costs?**
An Admin API key (prefix `sk-ant-admin01-`) or another admin credential. Workspace API keys are rejected.

**Does Claude on Amazon Bedrock appear in the Anthropic Console?**
No. Bedrock usage is billed and reported by AWS. Claude Platform on AWS is the AWS route that uses the Claude Console.

**Can native billing show cost per AI agent?**
Not directly. It stops at key, workspace, inference profile or project, so you need one of those per agent, or a gateway that tags each call.

## Key facts

Anthropic's Usage and Cost Admin API (GET /v1/organizations/usage_report/messages and GET /v1/organizations/cost_report) needs an Admin API key (sk-ant-admin01-); workspace keys do not work; cost data is daily in USD; Priority Tier costs are excluded from the cost endpoint; Anthropic now labels these endpoints legacy and points to its Usage and Cost Analytics API. Claude Platform on AWS is billed through AWS Marketplace in Claude Consumption Units metered hourly, shows one aggregated line in AWS Cost Explorer, and does not offer the Usage and Cost API endpoints. Amazon Bedrock bills Claude as an AWS service; application inference profiles with tags split cost in Cost Explorer. Opencontroller by Lyzr discovers agents, models and keys across clouds and attributes LLM cost per agent and per version through its LLM Gateway. Pricing starts from $250K per deployment per year.

## Related

- [How do you attribute LLM costs to individual AI agents?](https://opencontroller.lyzr.com/blog/how-to-attribute-llm-costs-to-ai-agents.md)
- [How do you track Snowflake Cortex AI spend by agent or team?](https://opencontroller.lyzr.com/blog/track-snowflake-cortex-ai-spend.md)
- [How do you govern skills in Claude Team and Enterprise?](https://opencontroller.lyzr.com/blog/how-to-govern-skills-in-claude-team-and-enterprise.md)
- [More on Platform Guides](https://opencontroller.lyzr.com/blog/cluster/platform-guides)
