AI Briefing
KOSign in

Anthropic introduces message threads beta to cut Claude Code time to first token by 35%

·2026.11.04 11:07

Key point

The new API parameter reduces median time to first token by 35% for 500k to 1M token requests by sending only new messages per turn.

Details

  • Anthropic released a beta thread parameter for the Messages API, enabling clients to send only new messages and the previous response ID instead of the full conversation history.
  • Median time to first token drops by approximately 35% for requests between 500k and 1M tokens, with smaller payloads improving efficiency for computer-use agents.
  • The feature is already enabled by default for most Claude Code users and is available on the Claude API, AWS, Google Cloud, and Microsoft Foundry.
  • Token usage, pricing, and rate limits remain unchanged, as Claude still processes the full conversation context for each turn.
  • Threads are stored in memory and expire 5 minutes after the last request, requiring the client to resend the full conversation to start a new thread.
  • Limitations include a maximum of 4 cache_control breakpoints, prohibition of system or tools in continue requests, and ineligibility for ZDR or HIPAA compliance.
latencyanthropicclaude-codellm-infrastructureapi-updatemessage-threads
Read originalShare on Threads

This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.

Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.

Related

Claude BlogHow message threads reduce latency for long-running agentsRedditClaude Code ignores AGENTS.md when CLAUDE.md exists unless explicitly importedRedditClaude Code 2.1.312 Halts Commits When No-AI Policies Are Present in ReposPyTorchKRClaude Code team doubles merges and cuts reverts by 40% with AI review systemCursorCursor removes malicious Claude Code extension version 2.1.313 after SSH backdoor attemptHacker NewsClaude Code background agent status updates consume 15-20% of tokens