Anthropic has launched a public beta of “message threads” on the Claude platform. This new feature is designed to reduce response latency in long-running AI agent sessions by eliminating the need to resend the entire conversation history with every API request. With message threads, developers can simply provide new messages along with the IDs of previous responses to continue the interaction. Anthropic confirmed that this feature is enabled by default for most Claude Code users.
Message Threads Now Available in Public Beta on the Claude Platform
In a blog post on Monday, Anthropic announced the public beta release of message threads on the Claude platform. The feature aims to lower latency during extended sessions; the company notes that with message threads, there is no longer a need to resend the full conversation history with every exchange.
The message threading feature is now available in public beta on the native Claude platform. This functionality also extends to the Claude platform on Amazon Web Services (AWS), Claude within Microsoft Foundry, and Google Cloud Claude models managed by Anthropic. Anthropic has confirmed that existing token pricing and API rate limits for Claude will remain in effect.
Coding agents may issue hundreds of requests within a single session; with the newly introduced message threading feature, developers only need to submit the latest messages added since the previous turn. This reduces the time to first token (TTFT).
According to Anthropic, for most Claude Code users, the default message threading configuration reduces average response latency by approximately 35% for requests containing 500,000 to 1 million tokens.
Developers can enable this feature by including a thread parameter in their Messages API requests. The initial request contains the full conversation and establishes a thread. Subsequent requests include only the new messages and the ID of the previous response. Anthropic notes that conversation threads are stored temporarily in memory between requests. A thread expires five minutes after the most recent request, at which point the client must resend the complete conversation to initiate a new thread.
The company also confirmed that Claude reads the full conversation history within each message thread; consequently, token usage remains unchanged. Anthropic notes that message threads are particularly well-suited for long-running coding agents, agents capable of using a computer, and workflows that are intensive in terms of documents and images.