Anthropic has released a beta Usage and Cost Analytics API that gives organizations a first-party way to retrieve detailed Claude usage and spending data. Announced on October 29, the interface adds separate endpoints for usage and cost reporting, with dimensions that include models, workspaces, API keys, service tiers and individual actors.
The release matters beyond finance teams exporting another report. Multi-model gateways, internal AI platforms and resellers often need to turn provider records into customer invoices, team chargebacks and margin reports. Anthropic’s new API provides more of the underlying data needed to do that without relying exclusively on token counts reconstructed from inference responses.
What Anthropic released
The beta exposes two organization-level GET endpoints: /v1/organizations/analytics/usage for consumption records and /v1/organizations/analytics/cost for financial data. Requests must include the beta header analytics-usage-cost-2026-08-24.
Organizations can filter or group results across several operational dimensions. Anthropic’s documentation lists model, workspace, API key, service tier, user and service account among the available fields. Reports can include token and request counts as well as actual and list-price costs.
That distinction between actual and list-price cost is particularly useful when contracted rates, discounts or other account-specific economics make a simple public-price calculation inaccurate. It can also help an operator separate the provider expense it incurred from the reference price used in internal reports or customer-facing billing.
The API is in beta, so it should not yet be treated as an immutable accounting contract. Integrations need to pin the required header, validate returned totals against existing invoices and tolerate changes consistent with a beta service.
Why gateways and resellers should care
A gateway can record the model requested, tokens reported and customer responsible for each call. That local ledger remains essential, but it may not fully explain the provider’s final charge. Service tiers, account-specific pricing and provider-side adjustments can produce differences between an estimated cost and the amount ultimately billed.
Anthropic’s cost endpoint creates an additional reconciliation source. A platform can compare its request ledger with organization-level provider records, investigate discrepancies and calculate margins using actual cost data. For an AI API reseller operation, that can improve customer chargeback and make unexpected cost movements easier to isolate.
The API-key and actor dimensions are also operationally significant. They allow reports to be divided among credentials, users and service accounts instead of stopping at a workspace-wide total. Organizations can use those divisions to identify unusually expensive keys, attribute shared Claude spending or support internal budget owners with more precise reports.
This does not eliminate the need for gateway-side metering. Provider analytics describe Anthropic consumption, while a multi-provider platform must preserve a consistent record across vendors and retain the customer, project or product context known only to the gateway. The stronger design is reconciliation between those ledgers, not replacement of one by the other.
Legacy integrations now face a migration decision
Anthropic has marked its earlier Usage and Cost API and several Claude Enterprise reporting endpoints as legacy. The company says those interfaces remain supported, so this is not an immediate shutdown notice. Existing integrations do not need to be removed simply because the beta exists.
Teams should nevertheless compare the new API with their current exports before choosing a migration date. Pagination, data freshness and grouping behavior can affect downstream jobs even when two interfaces report broadly similar information. A change that looks like an endpoint substitution may require updates to schemas, aggregation logic and historical comparisons.
Running both paths temporarily would let an organization compare totals by day, model and workspace. Material differences should be understood before the beta data becomes the basis for invoices, customer statements or automated AI API spend controls.
A higher standard for multi-model reporting
The release also raises expectations for gateway analytics. Customers increasingly need more than a token total: they need requested and charged models, service tiers, credential ownership, provider cost and the price presented to the customer. First-party reporting dimensions make omissions in a gateway’s own ledger more visible.
For platforms such as Model Gate, the practical opportunity is to use provider analytics as a reconciliation input within unified AI API billing. The provider feed can verify Claude expenses, while the gateway continues to normalize usage across vendors and associate it with teams, customers and managed API keys.
What remains uncertain is how stable the beta contract will be and how closely its records will align with final billing across different account arrangements. Until those questions are tested in production, the new endpoints are best treated as a valuable accounting source with reconciliation controls around them, not as the sole ledger of record.