Anthropic Launches Message Threads for Long Agent Sessions
Anthropic launched Message Threads in public beta on the Messages API on Nov 2: the first request sends the full conversation, and later turns send only new messages plus the previous response ID, with the thread held in server memory.

What happened
Until now, an agent resent every earlier message and tool result on each turn, and once a conversation reached hundreds of thousands of tokens, the resend itself slowed down the model's first response. Anthropic's Message Threads, launched Nov 2, targets exactly this step: the first request sends the full conversation and opens a thread, and later turns send only new messages plus the previous response ID.
Anthropic says the feature is already on by default for most Claude Code users, and on requests of 500,000 to 1 million tokens, median time to first token falls by about 35%. In long conversations, the median request shrinks from about 800 KB to 1.8 KB, because history is no longer re-uploaded.
Threads are held in server memory, encrypted and never written to disk, and expire 5 minutes after the last request, after which the client resends the full conversation. Anthropic notes the model still reads the full conversation each turn, so token usage, pricing and rate limits are unchanged; the feature is not available to zero-data-retention or HIPAA organizations or those with inference hooks enabled.
Key facts
- Launch and status: Anthropic, Nov 2, 2026, Message Threads in public beta on the Messages API
- How it works: the first request sends the full conversation, system prompt and tools with the thread parameter; later requests send only new messages and the previous response ID
- Latency effect: median time to first token falls about 35% on 500K–1M-token requests; the median request in long conversations drops from about 800 KB to 1.8 KB
- Retention: threads live only in encrypted memory, never on disk, expire after 5 minutes of inactivity, and have no list, read or delete API
- Platforms: the native Claude Platform, Claude Platform on AWS, Claude in Microsoft Foundry, and Claude Models on Google Cloud under Anthropic management
- Billing and limits: token usage and pricing are unchanged; Message Batches, on-demand compaction and most context editing are not supported
Why it matters
resending the full conversation on every long-session turn is now handled by a server-side cache for the first time, cutting time to first token by about 35% while billing and usage rules stay unchanged, so the latency bottleneck for long agent sessions shifts from transfer to the model itself.
CompaniesAnthropic
Follow HIMIAO Horizon
No sign-up · Unsubscribe any timeNotifications
A ping on your phone or computer when the brief is out. No email needed.
Comments
Related
- Oct 27Anthropic Merges Claude Chat and Cowork Into One Enterprise Conversation
- Oct 22Anthropic Launches Claude Duet, a Standalone AI Agent App
- Oct 12India's HCLTech Signs Pact With Anthropic, Sets Up Dedicated Unit for Enterprise Agents
- Oct 8Google Launches Universal Work Agent Gemini Agent, Running Gemini and Anthropic's Claude
- Oct 7Anthropic Expands Claude for Startups: A Free Year of Team, $1,000 in API Credits
Today
Nov 4 Wednesday- Brief
- GitHubGitHub Weekly Vol.4: 6 Projects Worth a Look This Week
Tested or read — weeklyAll →
- Quiz
- Call
Latest news
All →- Nov 4Fal Raises $1.3B Series F at $16B Valuation
- Nov 4OpenAI Raises Bug Bounties: Critical Finds Up to $30,000, Exceptional Finds up to $172,000
- Nov 4CrowdStrike Signs Two Deals in a Day to Buy Fiddler AI and Dreadnode
- Nov 3NYT Sues Brave Over AI Scraping of Its Content
- Nov 3SemiAnalysis Launches GDPx: Even the Best Model Scores 12%
- Nov 3Anthropic's Amended S-1: $16.96B Q3 Revenue, $895.4B in Compute Commitments
™