Anthropic message threads cut the wait on long Claude Code sessions

On 2 November 2026, Anthropic's platform release notes added message threads to the Messages API, in beta, under the header message-threads-2026-08-12. The notes say a long conversation can now send a smaller request and get a lower time to first token, because the API keeps the conversation on the server and later requests send only the new messages.

The product post, Make long-running agents faster with message threads, says the feature is on by default for most Claude Code users. On Claude Code requests of 500,000 to 1 million tokens, median time to first token drops by about 35%.

What actually changes

Without a thread, every turn of a coding agent uploads the whole conversation again, including earlier tool results. The docs say the API then validates and tokenizes all of it again. With a thread, the first request sends the full conversation, the system prompt, and the tools, and sets "thread": {"type": "create"}. Each later request sends only the new messages and the id of the previous response. System and tools stay as they were on the create call.

The article is explicit about the bill. Each request is smaller, but Claude still reads the full conversation on every turn, so token usage stays the same. Standard token rates and existing rate limits apply. The same beta is on the Claude API, Claude Platform on AWS, Claude in Microsoft Foundry, and Claude Models on Google Cloud when Anthropic manages the models.

The five-minute limit

A thread is not a saved chat. The article says each thread is held in memory, encrypted, and not written to disk. It expires five minutes after its last request. The client then resends the full conversation and starts a new thread. The docs add that the API deletes a thread no more than one hour after its last use, and that a continue call returns a 404 when the thread is gone. Your application is still the source of truth: the docs say no response tells you whether that turn was stored. You find out on the next continue.

The docs also separate this from session threads in Claude Managed Agents. Same word, different object.

The post lists where the latency cut matters most: long coding agents, where each tool result or file edit adds to the conversation; computer-use agents, where screenshots stay with the thread instead of being uploaded again; and document or image workflows. The more context the agent already carries, the more the first token moves up.

What this is for people waiting on an answer

The part a person feels in a long Claude Code session is the pause before the first token. Anthropic's own account is that the pause gets shorter on the biggest requests, and that the invoice does not. Idlen's advertisers page sells a native ad during that wait, in the editor or the browser, on CPM or CPC from €10. This beta does not add a new surface, and it does not change what that page prints.