Anthropic put message threads in public beta on the Claude Platform, announced November 2. An agent sends the conversation once, then sends only new messages and the previous response ID through a `thread` parameter on Messages API requests. Anthropic says that with threads on by default for most Claude Code users, requests of 500,000 to 1,000,000 tokens see median time to first token fall by about 35%. It publishes no chart, test setup or figure for smaller requests. Token use is unchanged. Claude still reads the whole conversation each turn, and standard rates and rate limits apply. The feature runs on the native Claude Platform, Claude Platform on AWS, Claude in Microsoft Foundry and Claude Models on Google Cloud. Anthropic's documentation says a thread is held in memory, encrypted and never written to durable storage. It expires 5 minutes after its last request and is deleted no more than 1 hour after last use. The same page lists the feature as not eligible for zero data retention and tells developers to keep their own copy of the conversation. For a team running coding or computer-use agents with very long sessions, this is a latency gain with no extra token charge. The cost case is neutral, and the speed case is Anthropic's own 35% median figure for requests of 500,000 to 1,000,000 tokens. Shorter sessions have less to gain, and Anthropic gives no number for them. The trade is where the conversation sits. Anthropic holds it for up to an hour after last use, and the feature is outside zero data retention. If your account relies on that agreement, check whether your tools turn threads on by default. The announcement does not say whether the Claude Code default applies to accounts with zero-retention terms. Check whether your Claude Code setup uses message threads and whether your data terms allow it. Anthropic's announcement gives one latency figure, a roughly 35% drop in median time to first token for 500,000 to 1,000,000 token requests, with no chart, methodology or test setup, and the documentation has none either. No visual could be verified, and the result has not been independently tested. Make long-running agents faster with message threads: Anthropic's announcement: the roughly 35% median time-to-first-token claim for 500,000 to 1,000,000 token requests, the 5-minute expiry, the in-memory and encrypted statement, and the unchanged token rates. Message threads, Claude API documentation: The developer reference: the beta header, supported models, the zero-data-retention eligibility note, deletion no more than 1 hour after last use, the 404 replay path and the fallback and compaction limits.