AI news for business · checked November 3, 2026

On November 2, 2026, Anthropic launched message threads in public beta: long-running agents now send only new messages instead of the full history.

On November 2, 2026, Anthropic released message threads in the Messages API as a public beta. Instead of uploading the full conversation history again with every request, applications only send the new messages and the ID of the last response. This is mainly meant to speed up long agent sessions, such as coding agents with many tool calls.

LOL.MARKETING / CONNECTING INTELLIGENCE

The essential answer

Message threads keep a Claude conversation on the server for a short time. The first request contains the whole conversation, after that only new messages follow. According to Anthropic, median time to first token drops by about 35 percent at 500,000 to 1 million tokens. Prices and rate limits stay the same, but the feature is not eligible for ZDR or HIPAA.

01 / FIT

A good fit when

  • Teams that run their own agents, coding assistants or computer use workflows with very long Claude conversations via the API.

02 / LIMITS

Not the first choice when

  • Companies with a zero data retention agreement or HIPAA readiness, and teams that only use Claude in the web app.

What are message threads?

According to Anthropic, a coding agent often makes hundreds of requests in one session, and until now every request resends all earlier messages and tool results. Once a conversation reaches hundreds of thousands of tokens, this resending alone adds time before Claude starts responding. Message threads address exactly this: agents only send what is new since the last turn.

The feature has been available as a public beta since November 2, 2026, on the Claude Platform, Claude Platform on AWS, Claude in Microsoft Foundry and Claude Models on Google Cloud (Anthropic managed). According to the documentation, it is in beta for every model. Message threads are different from the session threads in Claude Managed Agents.

How a thread works

As before, the first request contains the whole conversation including the system prompt and tools, plus the thread parameter with the type create. The API stores the conversation, including Claude's reply, under the ID of that response. On every later request, the application only sends the new messages and the type continue with the ID of the previous response. The system prompt and tools are then left out; the thread keeps them.

Every request in a thread needs the beta header message-threads-2026-08-12. A thread is best effort: it expires 5 minutes after its last request. If the API can no longer continue it, a 404 error is returned and the application sends the full conversation again in a new create request. The Anthropic SDKs include a thread helper that performs this replay automatically.

What it does – and what it does not

Anthropic names a concrete effect: message threads are already on by default for most Claude Code users, and for requests with 500,000 to 1 million tokens, median time to first token drops by about 35 percent. According to Anthropic, the more context an agent carries, the more the feature can reduce waiting time.

Important for cost planning: Claude still reads the full conversation on every turn, so token usage stays the same. According to the documentation, threads change neither pricing nor rate limits. Prompt caching is not automatic; anyone who wants caching still sets cache breakpoints or a top-level cache_control, with at most 4 breakpoints per request.

Data retention, ZDR and HIPAA

According to Anthropic, the API holds a thread encrypted in memory, never writes it to durable storage and deletes it no more than one hour after its last use. It stores the conversation sent to the model, the system prompt and tools from the create request, and a fingerprint of certain fixed fields. Files are only held by reference.

Message threads are explicitly not eligible for zero data retention (ZDR) or HIPAA. Organizations with HIPAA readiness enabled or with inference hooks turned on cannot use them. According to the documentation, using threads does not affect other ZDR eligible features. Anyone with a ZDR agreement should therefore align with data protection and IT before using threads.

Limits and next steps for teams

According to the documentation, threads do not support Message Batches, token counting, on-demand compaction, and context editing only with one exception. There is no API to list, read or delete threads. A thread is append-only; anyone who wants to edit or summarize earlier messages starts a new thread. The application itself therefore remains the source of truth for the conversation.

For companies with their own Claude agents, a short test is worthwhile: measure the response time of long sessions with and without a thread, build the replay path for 404 errors in from the start and check whether your data protection rules allow short-term server storage. Since the feature is still in beta, it should first run in non-critical workflows.

Keep it verifiable

Primary sources

The next sensible step

Are your Claude agents fast and compliant?

Eight steps from a general interest in AI to a clearer decision for your business.

Start AI System Check
FreeProvider-neutralNo credentials