Anthropic Lightens the Load for Long-Session Agents: Message Threads in Public Beta Cut Time to First Token by About 35%
Anthropic's official post announces a public beta of message threads: after the first turn, long sessions send only new messages, cutting median time to first token by about 35% on requests of 500,000 to 1 million tokens, with token usage and billing unchanged.
- 01THE ANNOUNCEMENT — Anthropic's official post on November 2 announces that message threads have entered public beta.
- 02THE CONFIRMATION — Independent reports such as Unite.AI verify the mechanism and figures, consistent with the official account.
- 03THE WATCHPOINT — When the beta limitations are relaxed, and whether threads move toward searchability and persistence.
WHAT HAPPENEDWhat Happened
Anthropic's official post says message threads have entered public beta on the Claude platform: an agent sends the full conversation only in its first request, and in each later turn sends only the new messages plus the ID of the previous response, while the server holds the conversation between turns.
The benefit figures the company cites come from the default-on rollout in Claude Code: on requests of 500,000 to 1 million tokens, median time to first token fell by about 35%. Independent reports put the change in request size at a drop from about 800 KB to 1.8 KB for a typical request in a long conversation.
The mechanism has clear boundaries: threads exist only in memory, are encrypted and never written to disk, and expire five minutes after the last request, after which the client must resend the full conversation to start a new thread; Claude still reads the full conversation each turn, so token usage, billing and rate limits are unchanged.
Availability covers the native Claude platform, the Claude platform on AWS, Claude in Microsoft Foundry, and Anthropic-managed Claude models on Google Cloud, billed at standard token prices and existing rate limits.
Anthropic also lists beta limitations: threads are best-effort, there is no API to list, read or delete threads, message batches, token counting and most context editing are not supported, and organizations with zero data retention or HIPAA requirements are out of scope.
WHY IT MATTERSWhat It Saves Is Latency, Not the Token Bill
Message threads solve the most annoying part of long sessions: retransmitting hundreds of thousands of tokens of conversation every turn, spending time before the model has even started speaking. Sending only the new content gets the first token out faster, but the model still reads everything, and the bill is untouched.
For developers, this is a pure latency optimization, not a cost optimization. Treating a thread as a cache is the right use; treating it as conversation storage is wrong. Applications still need to keep their own complete records and retain a full-resend path.
WHY IT MATTERSInfrastructure for Long-Session Agents Is Being Filled In
A coding agent can issue hundreds of requests in a single session, and screenshots and tool results pile up, so retransmission costs grow linearly with the session. The arrival of mechanisms like message threads shows long sessions have moved from demos to everyday workloads, and platforms are starting to build infrastructure for them.
The next signal is when the beta limits are relaxed: the day threads become searchable and persistent and support zero data retention, the deployment shape of long-session agents will change again.
SIGNAL JUDGMENT — ANALYSISA Latency Optimization, Not a Cost Optimization
Tokenta Signal's judgment: the value of message threads lies in the responsiveness of long sessions, not in the bill. Use them as a cache, and keep your own conversation records in your own hands.
WHAT REMAINS UNCERTAINWhat Remains Uncertain
EVIDENCEEvidence
Confirmed by two sources, Anthropic's official post and independent Unite.AI reporting; the original link was opened and verified.
PRIMARY SOURCESPrimary Sources
- Anthropic official post
https://claude.com/resources/articles/message-threads - Unite.AI independent report
https://www.unite.ai/anthropic-launches-message-threads-beta-for-long-running-agents/
RELATEDRelated Intelligence
- Microsoft Winds Down Its Oracle AI Cloud Rental as Older Nvidia Chips Fall Behind
- An Unreleased Anthropic Model Proved an 18-Year-Old Quantum Conjecture, Machine-Checked in Lean
- IBM Unveils Quantum System Two in Chicago, Its Most Advanced Computer Yet, Powered by Nighthawk
- Vistra Posts $1.14B Q3 Profit as Power EBITDA Jumps Over 50%, Lifts 2026 Outlook