Claude gains Message Threads for long agent conversations
Prepared by Sintelligent with the help of SI. Published in SI: . Updated: .
Long agent sessions can start responding faster without changing cost: the network transmits only changes, although the model still considers the entire conversation.
In the public API beta, an agent sends the full history once and then sends only new messages and the ID of Claude's last reply instead of repeatedly transmitting the entire conversation. According to Anthropic, median time to first token on requests of 500,000 to 1 million tokens fell by about 35%. The model still reads the full history, so token usage, prices, and rate limits are unchanged. The encrypted thread is held only in memory and expires after five minutes of inactivity. The feature works with all Claude models on Anthropic's platform and through AWS, Microsoft Foundry, and Google Cloud; most Claude Code users have already been switched to it.
Sources 2