Anthropic billing totals are useful but incomplete. Opsmeter links each request to tenant and endpoint ownership so cost spikes become actionable quickly.
Teams can read provider invoices but struggle to explain where Anthropic spend is concentrated operationally.
Assign spend anomalies to the exact feature owner instead of debating totals in Slack.
Escalate warning and exceeded states with request evidence attached.
Detect cost/request drift after prompt or model changes before monthly close.
Keep externalRequestId, endpointTag, tenant, and user fields stable across retries.
Opsmeter harmonizes token/cost payload differences into one schema.
Sort by endpoint, tenant, and prompt version to find true cost drivers.
| Dimension row | Metric | Context |
|---|---|---|
| support.reply endpoint | $1,280 / month | Top cost endpoint with high retry pressure. |
| tenant: eu-enterprise | 38% of total spend | One enterprise tenant dominates usage mix. |
| promptVersion support_v9 | +17% output tokens | Post-release verbosity increase detected. |
Start with one endpoint, then expand attribution coverage by tenant, user, and workflow.