Long chats no longer resend full history: Claude message threads keep the conversation server-side, so later requests carry only new messages

Long conversations and agent loops can now send only the new messages instead of resending the whole history with every request, so requests are smaller and time to first token is lower. The prior condition was to send the full conversation history with each request, which grows with every turn of a long session or agent loop. The capability is in beta on the Messages API under the message-threads-2026-08-12 beta header, and it is available on the Claude API, Claude Platform on AWS, Claude Models on Google Cloud (Anthropic managed), and Claude in Microsoft Foundry.

The API holds the conversation on the server, and each later request carries only the new messages. Message threads are the mechanism that carries this change, replacing client-side replay of the history on every turn. A thread expires 5 minutes after its last request and is deleted no more than 1 hour after its last use, which sets the window in which the server keeps the conversation.

The feature is in beta and requires the message-threads-2026-08-12 beta header, 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 suits continuously active sessions rather than server-side conversations that must persist long term.

The release note gives no specific figures for the improvement in time to first token or request size, and it does not say how a client should recover a conversation after a thread expires. The platform list reflects availability at release, and support on other platforms is not stated.