Anthropic's November 2 release notes introduce message threads in the Messages API. The company says sending only the new turn reduces request uploads and time to first token as an agent conversation grows. This is an API feature, not a new Claude model.

Keep a replay path

A thread expires after five idle minutes, and Anthropic says it deletes the retained conversation within one hour of last use. Applications must keep their own message history and be ready to resend it when a thread cannot continue. The SDK helper can handle that replay, but its own history is only in memory unless the application saves it.

Latency is separate from token cost

The documentation says threads do not change pricing or rate limits. They are beta and require the message-threads-2026-08-12 header; its embedded date is not the November announcement date. The feature is not eligible for zero data retention or HIPAA workloads. We checked the documentation, not a live performance test.