What happened
Anthropic launched the public beta of message threads on November 2, 2026. This new Messages API feature allows developers to stop resending entire conversation histories on every turn. For Claude Code users, the feature is already live by default, reducing median time to first token (TTFT) by approximately 35% for requests between 500,000 and 1 million tokens. The system caches conversation state in server memory for up to one hour, requiring developers to implement replay logic for expired threads.
Anthropic opened the public beta of message threads on November 2, 2026. This feature is part of the Messages API and allows programs to avoid resending the entire conversation history on every turn. Instead, the API caches the conversation on the server between turns, so clients only send new messages and the previous message ID.
For Claude Code, the feature is already live for most users without requiring a specific client update. According to launch figures attributed to Anthropic's internal testing, median time to first token (TTFT) drops by about 35% for requests between 500,000 and 1 million tokens. In long sessions, the median request size shrinks from roughly 800 KB to 1.8 KB, reducing upload data by more than 400 times.
The feature is available on the native Claude API, Claude Platform on AWS, Claude in Microsoft Foundry, and Claude models on Google Cloud managed by Anthropic. Amazon Bedrock is not currently on the official list. Developers must include the message-threads-2026-08-12 beta header in every request. Official SDKs for Python, TypeScript, Go, Java, C#, PHP, and Ruby include helpers to manage thread creation, continuation, and replay.
Message threads are append-only, and each response ID can be continued only once. The application remains the source of truth for the conversation and must maintain its own copy of every message, as no response confirms the turn was stored. If a thread expires, a continue request returns a 404 error, requiring the client to resend the full history to start a new thread.
The feature does not change pricing or rate limits. Token consumption remains identical because Claude still rereads the entire conversation on every turn. The primary benefit is speed, not cost savings. Prompt caching can be used alongside threads to reduce costs, but threads alone do not affect the bill.
Source details: pasqualepillitteri.it
Why it matters
This update addresses a critical performance bottleneck in long-running sessions, where re-uploading massive context windows significantly delays responses. By shifting the burden of state management to the server, Anthropic improves the practical usability of Claude Code for complex, multi-hour coding tasks. The feature does not change pricing or rate limits, meaning it is a pure latency optimization rather than a cost reduction. This distinction is crucial for enterprises managing API budgets, as it improves developer experience without altering the economic model of token consumption. The short-lived, in-memory nature of the cache also reflects a privacy-focused design choice, ensuring no persistent data storage on Anthropic's infrastructure.
Long-running coding agents often generate conversations that exceed hundreds of thousands of tokens. Without message threads, the API must receive, validate, and tokenize the entire conversation from scratch on every request, adding significant delay. Message threads eliminate this overhead by keeping the state on the server, allowing Claude to start responding sooner.
The short-lived, in-memory nature of the cache (up to one hour) is a deliberate design choice that prioritizes privacy and security. Unlike competitors who may store conversation state for weeks, Anthropic deletes threads from memory shortly after last use, with no data written to disk. This approach aligns with enterprise concerns about data retention and compliance.
While the feature improves latency, it does not reduce the cost of API usage. Developers must still pay for all tokens processed, including those in the cached conversation. This means the feature is most beneficial for users who are sensitive to latency but not necessarily to cost, such as those running interactive coding sessions where immediate feedback is crucial.
The requirement for clients to implement replay logic adds engineering complexity. Developers must ensure their applications can handle thread expiration by resending the full conversation history. This is a trade-off for the privacy and performance benefits of the feature, and it may be a barrier for some teams with limited resources.
Interactive Mechanism: How It Actually Works
Explore the underlying technology behind this development interactively.
crm_get_transaction(id='4092').An agent must create a draft calendar event for Tuesday at 2 p.m. Which evidence would establish the requested result?
What to watch next
Developers should monitor the stability of the beta header (message-threads-2026-08-12) and ensure their applications handle 404 errors for expired threads. Enterprises should evaluate whether the latency gains justify the engineering effort required to implement conversation replay logic. Additionally, watch for potential expansion of this feature to other API endpoints or models, and observe how competitors like OpenAI and Google respond to this specific approach to state management.
Monitor the stability and performance of the message threads beta, particularly in high-load environments. Watch for any changes to the beta header or SDK helpers that may require code updates.
Evaluate the impact of the feature on your specific use cases. If your applications involve long-running agents with large context windows, the latency improvements may be significant. For shorter interactions, the benefit may be minimal.
Keep an eye on how competitors respond to this feature. OpenAI and Google have similar mechanisms for managing conversation state, but with different retention policies and cost implications. Anthropic's approach may influence industry standards for state management in AI APIs.
Watch for potential expansion of message threads to other API endpoints or models. Currently, the feature is available for all Claude models, but it is not supported on Amazon Bedrock. Future updates may extend support to additional platforms.