Anthropic Ships Message Threads: What It Means for Agents That Read Your Form Answers (2026)

Message threads are the newest beta in the Claude API. Anthropic added them to its API release notes on November 2. With a thread, the API keeps a conversation on the server for a short time, so each later request sends only the new messages. The full guide is in the message threads documentation.

It sounds like plumbing, and it is. But it fixes a cost every long-running agent pays on every turn. Here is what shipped, what it does not change, and why it matters for agents that work through form answers.

What Anthropic shipped

Without a thread, every turn of an agent loop uploads the whole conversation again. The API validates and tokenizes all of it, turn after turn. So the longer the loop runs, the slower each answer starts.

With message threads, you send the full conversation once, with your system prompt and tools, and add a thread parameter set to create. On each later turn, you send only the new messages and the id of the previous response. The API rebuilds the conversation before it calls the model. As a result, Anthropic says time to first token "stays lower and steadier as the conversation grows".

A few details from the documentation:

• It is a beta. Requests need the message-threads-2026-08-12 beta header.
• It is short-lived. A thread expires 5 minutes after its last request, and the API deletes it no more than 1 hour after its last use.
• It works on the Claude API, Claude Platform on AWS, Claude Models on Google Cloud and Microsoft Foundry, all in beta.
• Pricing and rate limits do not change. This is about speed, not a discount.

What message threads do not change

Your application still owns the conversation. The documentation is direct about it: keep your own copy of every message and build a replay path from the start. A thread is best-effort, so if the API can no longer continue it, the request fails and you send the full conversation again. The SDK helper handles that replay for you.

There is a data point too. A thread's contents are held in memory, encrypted, never written to durable storage, and deleted within an hour. However, the feature is not eligible for zero data retention, so teams with that requirement should read the data handling section first.

What it means for agents that read form answers

Think about a common agent job: reading this week's form responses one by one. The agent lists them, opens each answer, scores the open text against a rubric, and writes a summary for the team. That is a long loop with a growing conversation, which is exactly the case message threads target.

With a thread, each step of that loop sends only the new response, not the whole history. So the agent starts each answer faster, and a review of 50 or 100 responses feels less like waiting.

dForms is the form builder for AI agents, so this kind of loop is the normal way to use it. Your agent builds the form, publishes it, and then reads every response with its score. Faster loops mean the reading part, where most of the agent's time goes, gets quicker. For the setup in Claude itself, see how to create a form with Claude. And if you let your agent publish on its own, our note on Claude auto mode covers who approves what.

Try it with dForms

Connect your agent to dForms (the connector at https://mcp.dforms.ai/mcp in Claude, or an API key for your own agent loop) and give it a long review task:

List every response to "Demo request" from the last 7 days.
For each one, read the open answer about what they want to automate
and score it 0 to 10 against this rubric: concrete process = 10,
vague goal = 4, blank = 0. Then give me the top 10 by total score,
with one line each on why they look like a fit.

In your own loop on the Claude API, add the thread parameter and the beta header, and keep your copy of the messages as the documentation asks.

Frequently asked questions

What are message threads in the Claude API?

Message threads keep a conversation on Anthropic's servers for a short time, so each request after the first sends only the new messages. The goal is a lower, steadier time to first token in long agent loops.

Do message threads make the Claude API cheaper?

No. According to Anthropic's documentation, threads do not change pricing or rate limits. Prompt caching is still how you lower the cost of a long conversation.

Do I still need to store the conversation myself?

Yes. Anthropic says your application stays the source of truth. A thread can expire, so keep every message and be ready to send the full conversation again.