Skip to content
Breaking:

Anthropic Adds Message Threads Beta to Claude API to Cut Agent Latency

The feature allows multi-turn agents to avoid retransmitting full interaction histories, cutting response delays on heavy context windows by an estimated 35%.

By The Company Wire3 min read
Share

Anthropic has launched message threads in public beta for its Claude API, introducing an architecture update aimed at speeding up multi-turn interactions for autonomous software agents. Reported by [The Next Web](https://thenextweb.com/news/anthropic-message-threads-beta-claude-api-agents), the feature removes the requirement for client applications to retransmit an entire conversation transcript on every successive request.

In typical agentic deployments—such as automated coding assistants running hundreds of tool calls and iterations—context histories frequently expand to hundreds of thousands of tokens. Under standard stateless API configurations, an application must pass the complete accumulation of previous messages, instructions, and environment outputs on each exchange, creating a cumulative transmission bottleneck before the language model can begin streaming a reply.

Under the new threads beta, an agent transmits full interaction history once during an initial call. For subsequent turns, the application submits only new user instructions or tool outputs alongside the unique identifier of Claude's prior response. According to Anthropic, bypassing full-payload retransmission reduces the initial response delay by roughly 35% on heavy sessions containing 500,000 to 1 million tokens.

The update is structured strictly as an operational latency optimization rather than an economic discount. Anthropic clarified that Claude still reads and processes the full conversation history on each turn, meaning per-token API prices and rate limits remain unchanged. To address the financial overhead of long agent sessions, the company had separately lowered prompt caching prices by 75% in September.

Data governance for the beta reflects temporary session handling. Anthropic maintains thread data strictly in volatile memory with encryption, stating that interaction records are never written to disk and automatically expire five minutes after the preceding API call. Developers must continue maintaining independent databases to retain permanent chat histories. The company also noted that threads are not currently compatible with its zero data retention policies or HIPAA-compliant healthcare deployments.

The beta functionality has been activated by default for most users of Anthropic's Claude Code command-line tool. It is accessible across all Claude model families through Anthropic's proprietary developer platform as well as third-party cloud infrastructure on Amazon Web Services, Microsoft Foundry, and Google Cloud.

Sources

  1. The Next Web

Company: Anthropic

Written by

The Company Wire

Newsroom · San Francisco

Inside the companies building what’s next. Reporting on startups, technology, funding and the people shaping them.