Anthropic publishes the design of its multi-agent research system
Published the design of its multi-agent research system — a lead agent runs subagents in parallel. In internal evaluation it outperformed a single agent by 90.2% and used about 15× the tokens of chat
Why it matters
Reached the opposite conclusion from the Cognition post above in the same month. Always cite the two as a pair
Source — Primary sources first; secondary reports are labelled “Press”
Officialanthropic.comhttps://www.anthropic.com/engineering/multi-agent-research-systemAdded in Week 35, 2026 · See that week