Message threads let a Claude agent send only its new messages on each turn, instead of resending the entire conversation every time. Anthropic put the feature into public beta on November 2, aimed at the lag that builds up in long agent loops. It is a practical speed upgrade for agents that run for dozens or hundreds of turns, but it comes with storage and compliance limits that should shape how you adopt it.
What Changes
Without a thread, every request uploads the whole conversation again, and the API validates and tokenizes all of it. With a thread, you send the full conversation once, then send only the new messages plus a reference to Claude's last reply, according to The Next Web.
The important caveat is that Claude still reads the whole conversation on every turn. Threads save time, not money. Prices and rate limits are unchanged. Anthropic says the gain shows up most in long sessions. On very long sessions of 500,000 to one million tokens in Claude Code, the typical wait before Claude starts answering fell by about 35%.
Where It Is Available
Anthropic's documentation lists message threads as a beta feature on the Claude API, Claude Platform on AWS, Claude Models on Google Cloud and Microsoft Foundry. The feature is tied to a beta header, message-threads-2026-08-12, which the documentation identifies as the feature's beta header. Check the current docs before deploying, because beta features change.
Setting It Up
1. Confirm support. Make sure the model and platform you use are on the supported list in the documentation. 2. Create the thread. Send your first request with the full conversation and the beta header, which creates the thread. 3. Continue the thread. For each later turn, send only the new messages and reference Claude's most recent reply. Follow the exact field names in Anthropic's guide, since third-party write-ups vary. 4. Keep your own copy. Store the full conversation on your side. Threads expire five minutes after the last request, according to The Next Web, and the documentation says they are deleted no more than one hour after last use. 5. Build a fallback. If a thread can no longer be continued, rebuild it from your stored copy by sending the full conversation again. Treat this as a normal code path, not an error case.
Limits to Know
Message threads are not eligible for zero data retention or HIPAA setups, according to both the documentation and The Next Web. The documentation says thread contents are held in memory, encrypted under a server-managed key, and never written to durable storage. That design may suit some workloads, but it does not satisfy strict retention rules, so compliance teams should review it first.
Because the feature is in beta, plan for changes to headers, field names and limits. Test the thread lifecycle in staging, including expiry, before you rely on it in production.
Why It Matters
For user-facing agents, latency compounds across turns. A response that begins a second sooner on a long session makes an agent feel more responsive, and a 35% cut in wait time on very long sessions is meaningful for coding and research agents that run long loops.
The compliance limits matter just as much. Teams handling regulated data cannot use threads yet, so message threads are a good fit for development tools, internal agents and workloads where data retention rules allow in-memory processing. Adopt the feature where the speed gain is real, and keep the full conversation stored on your own side so a thread failure is a retry, not a lost session.
