As a coding conversation grows, sending its entire history with every request can add delay. Anthropic’s public beta for Message Threads changes how that history is sent to its Messages API.
How a thread changes the request
The first request supplies the conversation, system instructions and tools. Later requests contain new messages and the previous response ID; the service keeps the thread briefly in memory. Anthropic reports about a 35% reduction in median time to first token for Claude Code requests containing 500,000–1 million tokens. That is a vendor measurement for a specific range, not a universal speed guarantee.
Expiry, pricing and data handling
Smaller requests do not mean lower token charges. Anthropic says the model still receives the full conversation and standard rates apply. A thread expires five minutes after its last request, so the application needs its own history and a replay path. The feature is not eligible for zero data retention or HIPAA.
Before adopting it, measure latency in your own long sessions and plan how your application will recover an expired thread.
