Anthropic Python SDK v1.15.0 adds speed parameter to Messages API and usage.speed field

Anthropic's Python SDK v1.15.0 adds a speed parameter to the Messages API, along with a matching usage.speed field in responses. The release was published by the anthropic-sdks-releaser account on October 15 at 19:45 (the release page does not show the year) and is tagged at commit bcd55fe.

This is an SDK release, not a model release. No new model, pricing, context window or benchmark data accompanies it.

New features

The release notes list three API-level additions:

• speed parameter and usage.speed on Messages. Requests can now pass a speed parameter, and responses report usage.speed. The release notes do not list the accepted values or explain how speed affects latency, output or billing. A documentation note in the same release says some models need a beta header to use speed. The notes do not say which models.
• between_tools added to thinking types in model capabilities. The model capabilities schema now includes between_tools as a thinking type. The notes give no definition. The name suggests it describes thinking that happens between tool calls, but that is an inference, not something the release states.
• Console usage-tier spend cap exposed as a read-only spend limit. The API now returns a Console organization's usage-tier spend cap as a read-only spend limit. Developers can read the cap programmatically, but the release does not indicate they can change it through the SDK.

Chores and documentation

The remaining changes are maintenance:

• CI: faster breaking-change detection.
• Docs: headers are now described on the Managed Agents and telemetry endpoints.
• Internal: the script that formats code blocks in docs is simplified.
• Tests: coverage added for OpenTelemetry options and for method default timeouts.

The docs item on Managed Agents endpoints and the OpenTelemetry test coverage show continued work on agent and observability surfaces in the SDK. The release notes do not describe either in detail.

What is not disclosed

The GitHub release page does not state:

• The allowed values or default for speed
• Which models require the beta header for speed
• Whether speed changes pricing
• A definition of between_tools thinking
• Whether any breaking changes are included (none are flagged in the notes)

Developers should check Anthropic's API documentation and the commit diff before relying on these fields in production.

What this means

The speed parameter is the most consequential change here. Its presence in both the request and the usage object suggests Anthropic is making a speed-related setting visible to developers, and that it can be metered or reported per request. The beta-header requirement indicates it is still gated and not available on all models. Until Anthropic publishes the details, teams should treat it as experimental and avoid assuming cost or latency effects.

The between_tools thinking type points to finer-grained descriptions of how models handle reasoning in tool-use workflows. Anyone building agents that choose models based on capability metadata should watch how this field is populated.

The read-only spend limit is a small but practical addition for teams that monitor organization budgets in code. It lets them check the usage-tier cap without reading it from the Console UI.

No migration is documented for this release. Teams on earlier 1.x versions should still review the diff, since the notes are terse.