Claude adds message threads for agents
Claude’s message-threads beta lets developers send only new messages after an initial request, reducing conversation-transfer overhead in long sessions.
Long-running agents accumulate a lot of conversation history. Sending that history again and again can add delay before a model even starts its next answer.
What happened
Anthropic introduced message threads in public beta on 2 November. Developers can send a conversation once, then submit new messages and the previous response identifier on later turns.
Anthropic measured an approximately 35% reduction in median time to first token for requests containing 500,000 to one million tokens among most Claude Code users. Token usage remains unchanged because the model still processes the conversation context.
Why it matters
The change targets the work of moving and handling a large request, rather than making the underlying reasoning free. That distinction matters for developers comparing response speed with total usage costs.
For agents that make many consecutive calls, less transfer overhead could make the application feel more responsive without changing its core workflow.
The bigger picture
Useful improvements to AI software often happen around the model itself. Request handling, state management and application infrastructure can influence the experience just as much as a new benchmark score.
Message threads are one example of that surrounding infrastructure. The announced latency figure applies to a specified class of long requests; teams will still need to measure the benefit for their own applications, conversation lengths and operating conditions.