---
title: "Anthropic's Message Threads cut long-conversation latency by 35%"
url: https://www.parallelquant.com/posts/anthropic-s-message-threads-cut-long-conversation-latency-by-35-c9a1b2
source_name: "The Decoder"
source_url: https://the-decoder.com/anthropics-message-threads-cut-time-to-first-token-by-35-percent-for-long-conversations/
published: 2026-11-03T12:35:49.000Z
topics: ["llms", "products"]
publisher: "Parallel Quant"
---

# Anthropic's Message Threads cut long-conversation latency by 35%

*2026-11-03 · Source: [The Decoder](https://the-decoder.com/anthropics-message-threads-cut-time-to-first-token-by-35-percent-for-long-conversations/)*

Anthropic released Message Threads for its Messages API. Instead of resending the whole conversation every turn, agents send only what changed. For requests of 500,000 to one million tokens, time to first token drops about 35 percent.

**Why it matters:** Long-running agents pay for resending huge contexts on every turn, in both latency and cost. Moving to stateful, incremental requests changes how agent loops are built and makes million-token sessions more practical. It also pushes API design toward server-side session state, a pattern other providers may follow.

**Topics:** llms, products

---
Read the original: https://the-decoder.com/anthropics-message-threads-cut-time-to-first-token-by-35-percent-for-long-conversations/
Canonical: https://www.parallelquant.com/posts/anthropic-s-message-threads-cut-long-conversation-latency-by-35-c9a1b2
Published by Parallel Quant — https://www.parallelquant.com