Claude Platform keeps conversations server-side: long chats and agent loops send only new messages, cutting time to first token

Claude Platform now offers message threads for long conversations and agent loops on the Messages API, so later requests carry only the new messages instead of the whole history. The release notes describe the change this way: the API keeps the conversation on the server, so each later request sends only the new messages. The capability ships in beta under the message-threads-2026-08-12 beta header and appears on the Claude API, Claude Platform on AWS, Claude Models on Google Cloud (Anthropic managed), and Claude in Microsoft Foundry.

Conversation state lives on the server, so the client appends new messages and requests become smaller with a lower time to first token. Message threads are a server-side session mechanism on the Messages API, turned on by a beta header rather than by client-side history management. A thread expires 5 minutes after its last request and is deleted no more than 1 hour after its last use, a lifecycle that applies across those four platforms.

The feature is in beta and requires the message-threads-2026-08-12 header in requests, and it is available on the Claude API, Claude Platform on AWS, Claude Models on Google Cloud (Anthropic managed), and Claude in Microsoft Foundry. A thread expires 5 minutes after its last request and is deleted no more than 1 hour after its last use, so it targets continuously active sessions rather than long-term conversation records.

The release notes give no measured improvement in time to first token or request size from message threads, and do not say how a client should restore context after a thread expires.