Message threads are in public beta on the Claude Platform. An agent no longer resends the full conversation history on every turn. It sends only what is new since the last turn, which shortens the wait before a long session starts responding.
[1]
The feature is on by default for most Claude Code users. Requests with 500,000 to 1 million tokens see a median time to first token drop of about 35 percent. A thread expires five minutes after its last request, and the client then resends the full conversation to start another. Each request is smaller, but the model still reads the full conversation, so token usage stays the same. Between turns, the thread is held in memory, encrypted, and not written to disk.
[1]The public beta covers the native Claude Platform, Claude Platform on AWS, Claude in Microsoft Foundry, and Anthropic-managed Claude models on Google Cloud. The page does not say what share of users leave the default off, and it does not break the 35 percent figure out by task.
[1]