Claude message threads: faster replies in long agent sessions

A new Messages API beta sends a conversation once, then only what is new, cutting wait time in very long sessions by about 35%.

On November 2, 2026, Anthropic opened a public beta of message threads in the Claude Messages API. Apps send a conversation once and then only the new messages, which Anthropic says cuts the median wait for the first token by about 35% in sessions of 500,000 to 1 million tokens.

Key facts

Anthropic announced message threads, a public beta feature of the Claude Messages API, on November 2, 2026.
After a first request with the full conversation, later requests carry only the new messages plus the ID of the previous response.
For requests of 500,000 to 1 million tokens, median time to first token drops by about 35%, according to Anthropic. Threads are on by default for most Claude Code users.
Token usage and prices do not change: Claude still reads the whole conversation on every turn.
The beta works on the Claude Platform, Claude Platform on AWS, Claude in Microsoft Foundry and Claude Models on Google Cloud (Anthropic managed).

What happened

Until now, every call to the Claude Messages API had to carry the entire conversation. In a short chat that is not a problem. In a coding agent that has been working for hours, with files, tool results and screenshots piling up, it means uploading a huge payload on every single turn, and the API has to receive and check all of it again before it starts answering.

Message threads change that. The first request sends the full conversation and opens a thread. From then on, the app sends only what is new, together with the ID of the last response. Anthropic says the result is a lower and steadier wait for the first token as the session grows, and it quotes a drop of about 35% in median time to first token for requests between 500,000 and 1 million tokens.

Threads are kept in memory, encrypted, and are not written to disk. A thread expires five minutes after its last request; after that, the app resends the full conversation and a new thread begins. Unite.AI, reading the documentation, adds that thread contents are deleted no more than one hour after last use, that threads are append only, and that changing the system prompt or the tools requires a new thread. Apps must still keep their own copy of every message, because there is no way to read a thread back.

Why it matters

Long-running agents are where waiting hurts most. If you use Claude Code or a similar tool, a few seconds saved on every step add up over a working day, and Anthropic says most Claude Code users already have threads switched on. You do not need to change anything to get that benefit.

For developers, the change is about speed, not cost. Claude still reads the full conversation each turn, so the token bill stays the same. To pay less for repeated content, Unite.AI notes, threads have to be combined with prompt caching, which threads do not apply automatically.

There are also limits worth knowing. Threads are not eligible for zero data retention or HIPAA use, and they do not work with Message Batches or token counting. Teams with strict data rules should check the documentation before turning the beta on.

How to try it

Threads make very long sessions faster, but they do not make them better. A chat that has grown huge can still drift, forget early decisions or repeat itself. Our free prompt “Move a long AI chat to a fresh one without losing context” helps you start clean when that happens. Paste it at the end of a long chat and fill in the fields:

What the chat is about: “Refactoring the login module of our web app with Claude Code”
What you want to do next: “Finish the password reset flow and write tests for it”
Details that must not be lost: “We use PostgreSQL, no new libraries, the API must stay backward compatible”

The answer should be a short brief with the decisions taken, the exact facts, the current state and the next steps. Before you paste it into a new chat, check that every file name, version number and decision in the brief really appears in your conversation. If something is missing or invented, correct it yourself. A good brief is short enough to read in a minute and complete enough that a new session does not need to ask you anything twice.

Try it

Rewrite text more clearly

Frequently asked questions

What are Claude message threads?

A public beta feature of the Claude Messages API, announced on November 2, 2026. After a first request with the full conversation, apps send only new messages and the ID of the previous response. The goal is a faster first token in long agent sessions, especially coding agents, computer use agents and document heavy work.

Do message threads make Claude cheaper?

No. Anthropic says token usage stays the same because Claude still reads the full conversation on every turn. Only the size of each request shrinks. Standard token rates and rate limits apply. To reduce costs for repeated content, developers need to combine threads with prompt caching.

How long does a message thread last?

A thread expires five minutes after its last request. Each new request restarts that window. When a thread has expired, the app sends the whole conversation again to start a new one, which is why apps must keep their own copy of every message.

Sources