Anthropic has introduced Message Threads, a new feature for its Claude Platform designed to reduce response delays in long-running AI agent sessions. Announced on November 2, 2026, the feature allows developers to maintain conversations without resending the entire conversation history with every API request.
The launch addresses an important challenge for AI agents: as conversations grow longer, requests become increasingly large because they contain previous messages, tool outputs and other context. Anthropic says Message Threads can reduce median time to first token by approximately 35% for requests containing 500,000 to 1 million tokens.
What Are Anthropic’s Message Threads?
Message Threads is a new capability in Anthropic’s Messages API that allows developers to maintain conversations on the server between consecutive requests.
Previously, developers building AI agents typically had to resend the complete conversation history whenever they wanted the model to respond. This included earlier user messages, Claude’s responses, tool calls and tool results.
For an AI coding agent working on a large software project, a single session could involve hundreds of interactions. Every new request would carry the accumulated conversation history, increasing the amount of information that needed to be transmitted and processed before Claude could begin generating its next response.
Message Threads changes this process by allowing developers to send the complete conversation only once. Subsequent requests need to include only newly added messages and the ID of the previous response.
This approach reduces the size of individual API requests and helps improve response speed, particularly in sessions involving large amounts of accumulated context.
Anthropic Reports 35% Lower Response Latency
One of the most important advantages of Message Threads is its impact on time to first token (TTFT).
TTFT measures how long an AI system takes to generate its first response token after receiving a request. It is an important performance metric because users and developers want AI agents to begin responding quickly, particularly when running complex tasks.
Anthropic says Message Threads reduces median TTFT by approximately 35% for requests containing between 500,000 and 1 million tokens.
The feature is already enabled by default for most Claude Code users, according to Anthropic.
| Metric | Details |
|---|---|
| Feature | Message Threads |
| Availability | Public beta |
| Launch announcement | November 2, 2026 |
| Latency improvement | Approximately 35% lower median TTFT |
| Tested context size | 500,000–1 million tokens |
| API requirement | thread parameter |
| Token pricing | Unchanged |
| API rate limits | Unchanged |
The improvement is particularly relevant for developers operating autonomous coding agents that repeatedly interact with files, execute commands, receive tool outputs and modify software projects.
How Message Threads Work
Anthropic has designed Message Threads to work through a relatively simple change to Messages API requests.
Developers need to add a thread parameter when creating or continuing a conversation.
The process involves three main stages:
1. Create a New Thread
The first API request includes the complete conversation, system instructions, available tools and a thread creation parameter.
Anthropic stores the conversation temporarily and associates it with the response ID.
2. Continue the Existing Conversation
For subsequent requests, developers no longer need to submit the entire conversation history.
Instead, they provide:
- Newly added user messages.
- The ID of the previous Claude response.
- The thread continuation parameter.
Claude can then continue the conversation using the context already maintained by the platform.
3. Handle Expired Threads
Message Threads are designed for active sessions rather than permanent conversation storage.
A thread expires five minutes after its last request. If a developer attempts to continue an expired thread, the application must send the complete conversation again to create a new thread.
Anthropic’s SDK helper can automatically handle certain thread recovery situations by replaying the conversation when continuation fails.
The feature therefore reduces repeated data transmission without eliminating the need for developers to maintain their own conversation history.
Message Threads Does Not Reduce Token Usage
Despite reducing request sizes and response latency, Message Threads does not reduce the number of tokens Claude processes.
Anthropic clarifies that the model still reads the full conversation on every turn. The feature primarily eliminates the need for developers to repeatedly transmit the complete conversation history through their API requests.
This distinction is important because smaller API requests do not necessarily mean lower AI inference costs.
Standard Claude token pricing and existing API rate limits remain unchanged. Developers should therefore view Message Threads primarily as a performance improvement rather than a cost-saving feature.
Developers looking to reduce token consumption will still need to use techniques such as prompt caching, context management and conversation summarisation where appropriate.
Where Message Threads Could Be Useful
Anthropic has identified several use cases where the feature could make AI agents more responsive.
Long-Running Coding Agents
Coding agents frequently interact with repositories, inspect files, execute commands and make changes to software.
These activities continuously add information to the conversation. Message Threads can reduce the overhead of transmitting this accumulated history with every new request.
This could make complex coding workflows smoother, particularly when an agent needs hundreds of interactions to complete a project.
Computer-Use Agents
Computer-use agents interact with graphical interfaces through screenshots, clicks and other operations.
As sessions become longer, screenshots and interaction results can create substantial conversation histories.
Message Threads allows developers to avoid repeatedly sending the entire history, potentially improving responsiveness in these workflows.
Document and Image-Heavy Workflows
AI applications that process large documents, images and other files can accumulate significant context over multiple interactions.
Message Threads can help reduce the size of subsequent API requests by retaining the existing conversation context temporarily.
This is useful for applications that need to perform several related operations on the same set of documents or visual inputs.
Availability and Developer Access
Message Threads is available in public beta through the native Claude Platform.
Anthropic has also made the feature available through:
- Claude Platform on Amazon Web Services (AWS).
- Claude in Microsoft Foundry.
- Anthropic-managed Claude models on Google Cloud.
Developers can activate the functionality by adding the thread parameter to Messages API requests and using the required beta header:
message-threads-2026-08-12
The feature is available across Claude models, subject to its beta limitations and supported platform configurations.
Anthropic also notes that Message Threads is not eligible for Zero Data Retention (ZDR) arrangements. Thread contents are held temporarily in encrypted server memory, are not written to durable storage and are deleted no more than one hour after their last use.
Developers handling sensitive information should review these data-handling conditions before enabling the feature.
The Bigger Picture
Anthropic’s Message Threads launch reflects a broader shift in AI development. The industry is moving beyond simple question-and-answer chatbots towards autonomous agents capable of completing complicated tasks over extended periods.
These agents need to remember previous interactions, use external tools and maintain context while working through multiple steps. As a result, managing conversation history efficiently is becoming an important part of AI infrastructure.
The 35% reduction in median time to first token for very large requests could make a meaningful difference in agent responsiveness. However, the unchanged token usage and pricing show that Anthropic is targeting performance and request efficiency rather than reducing the underlying computational cost of AI models.
Looking Ahead
Message Threads could become an important building block for developers creating coding assistants, computer-use agents and other applications that require long-running interactions. Its public beta availability gives developers an opportunity to test the feature and measure its impact on their existing AI workflows.
As AI agents become more capable and handle increasingly complex tasks, improvements in context management and response latency will likely become important competitive factors among AI infrastructure providers. Anthropic’s latest release represents another step towards making long-running AI agents more practical and responsive.
Get the day’s top stories in your inbox
One concise email. No spam, unsubscribe anytime.
