Anthropic Adds Message Threads to Speed Up Long-Running Claude Agents
A new Messages API beta lets agents send only what's new each turn, cutting time to first token without changing token usage.
Published
Last updated / Fact-checked on
Anthropic has launched message threads in public beta for its Messages API, letting agents send only new messages each turn instead of the full conversation, according to an Anthropic article. Anthropic says requests of 500,000 to 1 million tokens see median time to first token drop about 35%. Token usage and prices stay the same.
Photo: Dean Moriarty / Pixabay
Message threads at a glance
| What | thread parameter in the Messages API |
| Status | Public beta |
| Speed claim | ~35% lower median TTFT on 500k-1M token requests |
| Token usage | Unchanged; standard rates and rate limits |
| Storage | In memory, encrypted, not written to disk |
| Expiry | 5 minutes after the last request |
| Platforms | Claude Platform, AWS, Microsoft Foundry, Google Cloud |
| Source | Anthropic |
In typical agent workflows, a coding agent might make hundreds of requests in one session. Each request previously required resending every earlier message and tool result. Once a conversation reaches hundreds of thousands of tokens, resending all of it adds time before Claude starts responding, as described in the Anthropic article.
How message threads work
With message threads, agents send only what’s new since the last turn. To use the feature, developers add the thread parameter to Messages API requests, according to Anthropic. The first request includes the full conversation and starts a thread. Each later request includes only the new messages and the ID of the previous response.
Between turns, Anthropic holds each thread in memory, where it is encrypted and not written to disk. A thread expires 5 minutes after its last request. After that, the client resends the full conversation to start a new one, the Anthropic article states.
Claude still reads the full conversation on every turn. As a result, token usage stays the same. Standard Claude token rates and existing rate limits apply, so the feature does not cut token costs.
Speed gains and use cases
Anthropic reports that with message threads on by default for most Claude Code users, requests with 500k to 1M tokens see median time to first token (TTFT) drop by about 35%. The more context an agent carries, the more message threads can reduce time to first token, according to the article.
The feature is best for active, latency-sensitive sessions. These include long-running coding agents, where each tool result or file edit adds to the conversation. It also suits computer use agents, where screenshots stay with the thread instead of being re-uploaded on every turn. Document- and image-heavy workflows benefit as well, because requests are made smaller, the Anthropic article notes.
Availability
Message threads is available now in public beta on the native Claude Platform. It is also available on Claude Platform on AWS, Claude in Microsoft Foundry, and on Claude Models on Google Cloud (Anthropic managed), according to Anthropic. Developers can consult the documentation to get started.
Why it matters
This matters to readers building or using AI agents because long conversations and tool-heavy workflows often grow large enough that resending history becomes a noticeable delay. Message threads addresses that latency without changing token consumption or pricing. Teams running coding agents, computer-use agents, or document-intensive processes can see faster responses on large contexts while keeping the same cost structure. For sessions that stay active within the five-minute window, the approach keeps requests smaller after the initial turn. Once a thread expires, the system falls back to sending the full history again.
What to watch
- General availability. When the feature leaves beta.
- Longer expiry. Whether Anthropic extends the 5-minute thread window.
- Independent latency tests. Whether developers see the ~35% gain outside Claude Code.
- Rival APIs. Whether other model providers ship similar stateful features.
More coverage is on our Artificial Intelligence hub.
How we reported this
Details come from Anthropic's article, which carries no publication date; the 35% figure is Anthropic's own measurement for Claude Code users and was not independently tested.
Published and fact-checked November 3, 2026.
Sources
- Anthropic, "Make long-running agents faster with message threads," accessed Nov. 3, 2026: https://claude.com/resources/articles/message-threads
DailyMech used AI tools to help research and draft this article; it was fact-checked by our editors.