Key points
- Anthropic opened a public beta of Message Threads, which sends only new content per request instead of the full conversation history each time.
- Anthropic says first-token latency drops roughly 35 percent on very long prompts, in the range of half a million to a million tokens.
- Billing is unaffected because Claude still processes the whole history; a thread is held encrypted in memory and lapses after five idle minutes.
- Available through Anthropic's own platform plus the AWS, Microsoft Foundry and Google Cloud channels.
Why it matters for marketers: Teams running long-context agents or document-heavy workflows may get faster responses without paying less or more, so latency-sensitive automations are the first place to test.
Anthropic has opened a public beta of Message Threads for Claude. Previously, each API call had to carry the whole conversation so far. With the new feature, a request sends only what is new since the previous message, and the model picks up the rest from a stored thread.
According to Anthropic, the payoff is a roughly 35 percent shorter wait for the first token when prompts are extremely long, from about 500,000 up to one million tokens. The company says coding agents, computer-use agents and workflows that chew through many documents or images gain the most.
There is a limit on the savings. Token counts stay the same, because Claude still reads the full history each time, so this is a speed change rather than a cost change. The thread is kept encrypted in memory and expires once it has been idle for five minutes, which means long pauses will break the shortcut.
For marketing and content teams, the relevance is mostly in agent-style work: reviewing large content libraries, processing batches of creative assets, or running long research sessions. Faster first responses make those tools feel more usable, particularly when a person is waiting on the output.
Teams should check Anthropic's documentation and test the beta on their heaviest workflows. Watch how the five-minute window fits your usage patterns, and whether the feature stays free of pricing changes once it leaves beta.
Sources
This brief is our own summary. Read the full reporting at the source.
