Anthropic Speeds Up Long Claude Conversations by 35%

Anthropic has released message threads for its Messages API, the system behind Claude. Normally, an AI agent resends the entire conversation history with every request. With message threads, the agent sends only what changed since the last message. For conversations between 500,000 and 1,000,000 tokens (the chunks of text the model reads), the median time to the first word of a response drops by about 35%. Claude still reads the whole conversation at every step, so total token usage and price stay the same. The feature is on by default for most users of Claude Code, Anthropic's coding assistant, and it is also available through AWS, Microsoft Foundry, and Google Cloud.

This fix is built for agents that run for hours, not minutes: coding agents that edit files step by step, or agents that take screenshots and act on them. Before this, every step meant rereading the entire past conversation, so long sessions felt like a growing wait before the answer even started. The gain is in speed within one long session, not in price.