Anthropic adds Message Threads to cut Claude long-chat latency

Anthropic has released Message Threads as a public beta for its Claude platform, according to The Decoder. The new Messages API feature is aimed at long conversations where developers previously had to send the full chat history with every request, a pattern that can slow down agent-style apps.

For builders using Claude in coding assistants, computer-use agents, or document-heavy workflows, the practical benefit is lower waiting time before a response begins. Anthropic says token usage does not fall, however, because Claude still reads the complete conversation history.

What Message Threads change in the API

Before this release, Anthropic’s Messages API required each new request to include the complete conversation so far. Message Threads change that workflow by sending only what has changed since the last message, while keeping the thread available between requests.

Anthropic says this reduces time to first token by about 35% for requests between 500,000 and one million tokens. Time to first token is the delay before the model starts producing an answer, so the change should be most noticeable in products where users sit through long pauses before seeing any output.

The company says the thread is encrypted in memory between requests. It expires after five minutes of inactivity, which means developers should not treat it as a persistent long-term store for user sessions.

Who is likely to benefit most

Anthropic points to three main use cases: coding agents, computer-use agents, and workflows that handle many documents or images. Those are exactly the areas where conversation state can grow quickly as the model keeps track of files, instructions, tool calls, prior outputs, and user feedback.

For small teams building AI products, the update matters less as a new end-user feature and more as an infrastructure improvement. If an app repeatedly sends very large context to Claude, Message Threads may reduce the delay before the first token appears. That could make a research assistant, support tool, or internal operations agent feel more responsive without changing the visible interface.

Developers working on coding-agent products may want to compare this with other Claude platform changes around agent behavior, including Anthropic’s recent work to tighten Claude Code prompts. The common thread is that Anthropic is continuing to tune Claude for longer-running, tool-using workflows rather than only single-turn chat.

Not a token-cost reduction

The most important caveat is that Message Threads do not reduce token usage, according to Anthropic. Claude still processes the full conversation history, even if the API client does not resend all of it each time.

That distinction matters for businesses watching AI budgets. Message Threads may improve responsiveness for long conversations, but they should not be expected to lower token consumption on their own. Teams that already monitor Claude costs will still need to track usage separately; Anthropic has also added reporting tools for Claude spending, which may be relevant for organizations scaling API workloads.

In practice, this means the feature is best evaluated as a latency optimization. If your main pain point is slow first responses in very long threads, it is directly relevant. If your main pain point is total model spend, the source does not indicate a cost-saving effect.

Availability across Claude providers

Message Threads are available in public beta on the Claude platform. The Decoder reports that the feature also works through AWS, Microsoft Foundry, and Google Cloud.

That cross-platform availability matters because many businesses do not access Claude directly through Anthropic. Some use cloud marketplaces or existing enterprise infrastructure, and support across those channels makes the feature easier to test without changing an entire deployment path.

For creators, marketers, freelancers, and small businesses, the near-term impact will mostly be indirect. You may notice faster responses in Claude-powered apps that manage large projects, analyze big collections of documents, or operate as agents over multiple steps. But the benefit depends on whether the app’s developer implements Message Threads and whether the workflow uses large enough context to make the latency reduction meaningful.

The feature is also a reminder that AI product performance is not only about model quality. For long-context applications, how much state has to move between the app and the model can shape the user experience just as much as the answer itself.