Anthropic Message Threads cut first token wait by 35 percent
The speed up applies to requests of 500,000 to one million tokens, and token use does not change.
Anthropic released Message Threads as a public beta on its Claude platform. Until this time, each API request (a message from software to the model) had to send the full conversation again. At this time, only the change since the last message goes. Anthropic tells that this cuts the time to the first token by about 35 percent. This is for requests of 500,000 to one million tokens.
How it works
The thread stays encrypted in memory between requests. It expires after five minutes of inactivity. Anthropic tells that coding agents and tools for many documents or images get the most help. It works on:
The numbers
The 35 percent number is from Anthropic. The source gives no test by other persons. Token use does not change, because Claude continues to read the full history.
What is not known
The source gives no number for requests of less than 500,000 tokens. It gives no price change.
Insight: Three teams cut the first token wait and none cut the cost
Anthropic stops the resend of history. In October 2026, researchers showed TokenGate, which removes tokens, and GraphTok, which speeds the step that turns text into tokens. This reads as a race to speed up long agent sessions. A vendor can cut the wait without a lower cost. Look for a price change.
Sources
Posted
