Anthropic has altered how its infrastructure handles heavy API workloads, shifting away from retransmitting complete conversation histories. As the company announced, Anthropic launched Message Threads in public beta across its Claude platform, addressing the network overhead that slows down automated agent deployments.
Every API call previously required clients to resend the entire conversation backlog. Under the new mechanism, Message Threads transmits only the delta that changed since the preceding message. According to Anthropic, this change reduces the time to first token by approximately 35 percent on large context requests ranging from 500,000 to one million tokens. The performance gains concentrate around coding agents, computer-use agents, and multi-document or image processing pipelines.
The underlying economics of token consumption remain unchanged, because Claude continues to read the full context history on each run. The thread itself stays encrypted in memory between interactions and automatically expires after five minutes of inactivity. For enterprise customers orchestrating complex workflows, as the platform details, the update is available directly through the Claude platform as well as via integrations across AWS, Microsoft Foundry, and Google Cloud.
What’s behind it
Developers previously had to resend entire conversation histories with every API call, but Message Threads now keeps the context in memory and accepts only incremental deltas. This architecture directly reinforces the pricing model introduced in early September, when cache read rates dropped to $0.25 per million tokens to support heavy autonomous loops. The optimization cuts time to first token by 35% on contexts reaching 1 million tokens, while in-memory storage lasts exactly five minutes of inactivity.
Our stories: Claude Fable 5.1 Cache Discounts and TCO Inflation (September 3).