What changed
Anthropic shipped version 1.15.0 of the official Python SDK on 15 October 2026, according to the public CHANGELOG.md on GitHub. The release adds API support for a speed parameter on Messages, surfaces usage.speed in responses, extends model capability metadata with between_tools thinking types, and exposes a read only Console organization usage tier spend cap as a spend limit field. Version 1.14.0 (14 October 2026) in the same changelog added OpenTelemetry tracing hooks and Managed Agents telemetry destinations, showing a two day burst of operator focused SDK work.
Why it matters
Application teams no longer have to infer latency tiers only from model choice. A first class speed parameter means routing logic, cost dashboards, and SLA reports can treat speed as an explicit product decision rather than a hidden provider default. Platform engineers should expect finance and procurement to ask for per tier usage splits once usage.speed lands in production logs.
Who is affected
Backend engineers integrating Anthropic Messages, FinOps owners of Claude API spend, and observability teams wiring OpenTelemetry after the 1.14 tracing release. Compliance staff may also care about the read only spend cap field when reconciling Console limits against internal budgets.
What to do next
Pin SDK versions in staging, upgrade to 1.15.0, and add integration tests that assert your client sends and parses usage.speed without breaking existing telemetry. Document which user journeys may opt into faster tiers versus quality first tiers before product managers expose toggles.
What to watch
Anthropic documentation notes that some models require a beta header to use speed; track which model IDs your production keys can access with the parameter enabled.
Image: Anthropic
Sources
- Primary — Anthropic, anthropic-sdk-python CHANGELOG.md v1.15.0 (15 October 2026). Speed parameter, thinking metadata, spend cap field.
- Secondary — Anthropic, Release v1.14.0 notes (14 October 2026). OpenTelemetry tracing and Managed Agents telemetry endpoints.
Related in AI
Continue reading

