Overview
- Message Threads is available in public beta on Anthropic’s Claude Platform and is enabled by default for most Claude Code users.
- The API flow requires the first request to send the full conversation to start a thread and later requests to send only new messages plus the previous response ID using a thread parameter.
- Anthropic says the change cuts median response latency by about 35 percent for very large requests of roughly 500,000 to 1,000,000 tokens by reducing network transmission of full histories.
- Threads are held only in server memory, expire five minutes after the last activity, are not written to standard disk, and are fully deleted within an hour, and the feature is not available to accounts requiring strict ZDR or HIPAA controls.
- Token usage, API pricing, and rate limits do not change because Claude still reads the full conversation on every turn, so the update improves perceived speed but does not lower billing; developers should expect faster agent responsiveness for long-running coding and heavy document workflows.