Anthropic Launches Message Threads Beta to Speed Up Long Claude Agent Sessions

According to Anthropic, median time to first token falls by about 35% for requests of between 500,000 and one million tokens. Unite.AI reported that the beta launched on November 2 and that threads are already switched on by default for most Claude Code users.

The company says coding agents, computer-use agents and workflows that process large numbers of documents or images will benefit most. These long-running sessions build up very large contexts, and until now every turn meant sending all of that context again.

How It Works

The first request in a thread contains the whole conversation, along with the system prompt and tools, and the API stores it together with Claude's reply under a response ID. After that, each request carries only the messages added since the previous reply, plus that reply's ID.

The savings are in speed, not cost. Claude still reads the full conversation on every turn, so token usage is the same as before. Unite.AI reported that standard token rates and existing rate limits apply.

Threads are encrypted and held in memory between requests, and they expire after five minutes without activity, The Decoder reported. Once a thread expires, the application has to send the full conversation again to start a new one. Developers therefore still need to keep their own copy of every message.

Limits and Availability

According to Unite.AI's summary of Anthropic's documentation, thread contents are never written to durable storage and are deleted no more than an hour after last use. The beta is not eligible for zero data retention or HIPAA use. That matters for companies in healthcare and other regulated sectors that depend on those guarantees.

Threads only allow new messages to be added. Changing the system prompt or tools means starting a new thread. The feature does not work with Message Batches, token counting, on-demand compaction or most context editing, and it is not available to organisations that have Inference hooks enabled. Prompt caching is not automatic inside threads, so developers have to set cache breakpoints themselves.

Message Threads is available on Anthropic's own Claude platform and through Claude Platform on AWS, Claude in Microsoft Foundry and Claude models on Google Cloud.

Why It Matters

For companies building agents on Claude, delay before the first response grows with context length and is one of the main costs of long sessions. The beta targets that problem without changing how usage is billed. It comes as Anthropic competes with OpenAI and others for enterprise and developer spending on agentic tools, and as it prepares for a public listing.