Anthropic's Message Threads Beta Lets Claude Agents Stop Resending Whole Conversations, Saving Time but Not Money
Anthropic released message threads in beta on November 2. The Claude API keeps a conversation on the server, so long-running agents send only new messages on each request. Anthropic says the wait for a first token fell about 35% on requests of 500,000 to 1M tokens, but prices and rate limits are unchanged. Threads expire five minutes after the last request and are not eligible for zero data retention or HIPAA setups.
Anthropic's newest API feature is aimed at the slowest part of long agent sessions, not the most expensive one. On Monday, November 2, the company released message threads in beta for the Claude Messages API. Instead of sending the entire conversation with every request, an agent sends it once; the API keeps it on the server, and each later request carries only the new messages.
What changes for developers#
Threads are switched on with the message-threads-2026-08-12 beta header, according to Anthropic's release notes. A thread expires five minutes after its last request and is deleted no more than an hour after its last use. The feature is available on the Claude API, Claude Platform on AWS, Claude in Microsoft Foundry and Claude Models on Google Cloud (Anthropic managed), the Anthropic-run Google Cloud option that became generally available in October.
The point is latency. A coding agent can make hundreds of requests in one session, each carrying every earlier message and tool result, The Next Web explains. Anthropic told the outlet that on requests of 500,000 to 1 million tokens, the typical time before Claude starts replying fell by about 35%, and that it has already turned threads on for most users of Claude Code.
What does not change#
Claude still reads the whole conversation on every turn, so the feature saves time rather than money: Anthropic says prices and rate limits are unchanged. Per The Next Web, threads are held in memory, encrypted and never written to disk, so developers must keep their own copy of each conversation. They are not eligible for Anthropic's zero data retention or HIPAA arrangements, which matters for customers in regulated settings.
Limits#
- The 35% figure is Anthropic's own measurement, relayed by The Next Web; Anthropic has not published the test setup, and we have not measured it.
- The release notes do not say when threads will leave beta or whether pricing could change then.
- The Claude Code rollout and the data-handling details come from The Next Web's report rather than the release notes.