Anthropic launches message threads beta for Claude API agents
The company reports about a 35% reduction in the wait for replies on very long requests, with no change to pricing or rate limits.
Anthropic has introduced a public beta of message threads for developers using its Claude API to build AI agents. The feature lets agents submit a conversation history initially, then send subsequent additions rather than repeatedly transmitting the entire exchange, the company said.
Most Claude Code users already have threads enabled, according to Anthropic. For requests containing 500,000 to 1M tokens, the company reported that the typical delay before an answer begins dropped by about 35%. Coding sessions can involve hundreds of requests, with earlier messages and tool results otherwise sent again each time.
After the initial submission, agents provide new content alongside an identifier for Claude’s previous response. Claude still processes the complete conversation on each turn, so the change reduces waiting rather than costs. Anthropic said it has left pricing and rate limits unchanged.
According to Anthropic, threads remain encrypted in memory without being saved to disk and expire five minutes after the most recent request. Developers therefore need to retain their own conversation records. Threads do not qualify for the company’s zero data retention or HIPAA health-data arrangements.
Every Claude model supports the beta. Developers can access it through Anthropic’s platform, AWS, Microsoft Foundry or Google Cloud.
