Long conversations can make each new API request slower when the entire conversation history has to be sent again. Anthropic’s Message Threads public beta changes how that history is transmitted: it sends only what has changed since the previous message. The company says this cuts time to first token by about 35 percent for requests between 500,000 and one million tokens.
Sending changes instead of the full thread
Before Message Threads, every API request had to resend the conversation history. As a conversation grows, that repeated transfer can add delay before Claude starts responding.
With Message Threads, the API transmits the updates since the last message. This reduces the amount of conversation data that needs to be sent again, which Anthropic says can shorten the wait for the first token in very long requests.
The reported improvement applies to requests between 500,000 and one million tokens. It is a reduction in time to first token, the time before a response begins; the source does not say that the feature reduces the time needed to complete an entire response.
Who may benefit most
Anthropic says the feature is especially useful for coding agents, computer-use agents, and workflows that process many documents or images. These kinds of tasks can involve long-running conversations with a substantial history that must remain available as work continues.
For such workflows, transmitting only the latest changes can make repeated requests more efficient to start. The conversation context remains available to Claude, while the repeated transfer of unchanged history is avoided.
Faster starts do not mean fewer tokens
Message Threads does not change token usage. Claude still reads the full conversation history, even though each request transmits only the changes since the previous message.
That distinction matters when assessing the feature. Its stated benefit is faster starts for certain long requests, not a reduction in how much of the conversation Claude processes or in token consumption.
Thread storage and availability
Between requests, a thread stays encrypted in memory and expires after five minutes of inactivity. This defines how long the stored thread persists when no further request is made.
The public beta is available on the Claude platform and through AWS, Microsoft Foundry, and Google Cloud. Anthropic’s documentation provides further details for teams considering how Message Threads fits into their API workflows.