anthropic-sdk-python v2.4.0

Version 2.4.0 of Anthropic's Python SDK, the library developers use to call Anthropic's APIs from Python, is mainly about the Admin API, which organizations use to manage members, workspaces, keys and spending from code instead of by hand. The headline feature adds "GA methods for analytics, spend limits and RBAC": generally available rather than beta methods for usage and cost reports, for per-member spending caps, and for role-based access control, which governs what each member is allowed to do (release notes).

Spend limits show what these methods are for. Anthropic's documentation describes the Spend Limits API as a Claude Enterprise feature for setting a spend limit on each member, seeing whether a member's limit comes from a personal override, their group, their seat tier or the organization default, and approving or denying members' requests for a higher limit (Spend Limits API documentation). This release adds types for filtering spend limits by period and makes the spend_limit field nullable in the response to an approved increase request.

The release notes do not say how the new methods differ from the existing beta ones. As of November 2, the documentation still showed spend limits under the SDK's beta namespace (client.beta.organization.spend_limits) and described the Usage and Cost Analytics API as beta. The release also adds two session error types, tool_use_limit_reached_error and compaction_limit_reached_error, without describing when they are raised. The TypeScript SDK's sdk-v1.4.0, released the same day, carries the same API changes, and earlier Intel coverage covers the Admin API's arrival in the SDKs.