Anthropic opens Message Threads beta, 35% faster first token
TL;DR
- Message Threads cuts median time-to-first-token roughly 35% on 500,000-to-1,000,000-token requests, Anthropic reports, with pricing and rate limits unchanged.
- Threads live in encrypted memory for five minutes, never on disk, with no retrieval endpoint, so apps must keep their own conversation copy.
- Beta ships on every Claude model via the native platform, AWS, Microsoft Foundry and Google Cloud; zero-data-retention and HIPAA customers are excluded.
Anthropic opened Message Threads in public beta on November 2, letting agents built on the Messages API stop resending the full conversation on every turn and instead pass only the newest message plus the ID of Claude's last reply. The company's launch post reports that "requests with 500k to 1M tokens see median TTFT drop by about 35%," with pricing and rate limits unchanged.
The mechanics come with hard edges. Threads live in encrypted memory, never written to disk, and expire five minutes after the last request; call a stale thread ID and the API returns 404, so apps still have to keep their own transcript on hand. "Claude still reads the full conversation on every turn, so token usage stays the same," the launch notes add, meaning the gain is purely latency, not spend.
The feature ships on every current Claude model and across the native platform, AWS, Microsoft Foundry and Google Cloud, hidden behind a `message-threads-2026-08-12` beta header. Anthropic says it has already flipped threads on for most Claude Code users, while TheNextWeb notes the feature is not available to customers on zero-data-retention or HIPAA setups.
Anthropic names long-running coding agents, computer-use sessions dragging repeated screenshots, and document-heavy workflows as the intended targets. That lines up with a fourth Anthropic alert we've logged today, alongside S-1A disclosures and a fresh copyright suit in the tracker.
Originally reported by claude.com
Read the original article →Original headline: Anthropic Opens Message Threads Beta, Cuts Time-to-First-Token 35% on 500K-1M Token Requests