How we built our multi-agent research system
Original title:How we built our multi-agent research system
AI summary
Claude's multi-agent research system uses an orchestrator-worker architecture, in which a lead agent analyzes the user query and formulates a strategy, calling multiple specialized subagents in parallel. Internal evaluations show that a system with Claude Opus 4 as lead and Claude Sonnet 4 as subagents outperforms single-agent Claude Opus 4 by 90.2% on research tasks. The system excels at breadth-first queries that require exploring multiple independent directions at once, scaling parallel reasoning by allocating independent context windows. However, multi-agent systems consume about 15 times the tokens of ordinary chat, making them suitable for tasks whose value justifies the performance gain; they perform best on tasks requiring high parallelism, large information volume, or many tool calls.
Key Takeaways
- Claude's multi-agent research system uses an orchestrator-worker architecture,in which a lead…
- Internal evaluations show that a system with Claude Opus 4 as lead and Claude Sonnet 4 as…
- 2% on research tasks.
- The system excels at breadth-first queries that require exploring multiple independent directions…
- However,multi-agent systems consume about 15 times the tokens of ordinary chat,making them…
Quick Facts
- Published
- Updated
- Category
- Opinion
- AI score
- 73
Anthropic 把 Research 背后的多智能体架构、prompt 工程和踩坑经验全抖出来了,做 Agent 的同行可以直接抄作业,尤其是『token 用量解释 80% 性能方差』这个结论,选型逻辑要变。
Related entities
Source & copyright
This article is based on curated titles and AI summaries from the AIHOT public API, with attribution and links back. Original copyright remains with the original publishers and authors; this site does not mirror full texts.
How to Cite
When citing this article, keep the title, date, and permalink, and credit Orange News as the source.
Orange News: How we built our multi-agent research system. June 13, 2025. https://new.nat6.net/en/news/cmorb7ml1008oslhfira08k2z/