Claude's API now lets agents skip re-sending the whole chat

Anthropic opened a public beta of message threads for its Messages API on November 2. Until now, an agent had to upload the whole conversation again on every turn. With a thread, the first request sends everything. Each later request sends only the new messages plus the ID of the previous response.

Anthropic says the wait before a reply starts stays shorter and steadier as a conversation grows. For requests of 500k to 1M tokens, it reports that the median time to first token fell about 35%, and it ties that figure to threads being on by default for most Claude Code users.

The feature is aimed at long agent loops: coding agents, computer-use agents that pass screenshots, and workflows heavy on documents or images. It needs a beta header. The docs list it on the Claude API, Claude Platform on AWS, Microsoft Foundry and Claude Models on Google Cloud.

The catch: Token usage does not change, because Claude still reads the full conversation each turn, so this saves time, not money. A thread expires five minutes after its last request, and the docs say threads are not eligible for zero data retention or HIPAA setups.

What it means if you build with Claude

• A coding agent with a long history could feel snappier, since less data goes over the wire each turn.
• Your bill stays the same, so don’t expect it to shrink.
• Teams under strict data-retention rules should check the eligibility notes first.

Key terms

• Time to first token: how long you wait before the model starts answering.
• Beta header: a setting you add to API requests to switch on a feature that is still being tested.
• Zero data retention: an arrangement where the provider keeps no copy of your requests after answering.