Stacks/Research
Anthropic's multi-agent Research system
Anthropic's Research feature is a production multi-agent system. A lead agent plans a research process and creates parallel subagents that search independently, then a citation agent grounds the final report. The company reports that this architecture beat a single Opus 4 agent by 90.2% on an internal research eval, especially on breadth-first queries.
Jeremy Hadfield, Barry Zhang, Kenneth Lien, Florian Scholz, Jeremy Fox, and Daniel FordJun 12, 2025Original writeup ↗
- Use case
- Open-ended research across the web, Google Workspace, and other integrations for Claude's Research feature
- Category
- Research
- Team size
- not stated
- Orchestrator / framework
- A custom orchestrator-worker architecture: a LeadResearcher plans the work, saves the plan, and spawns specialized subagents that search in parallel; a CitationAgent then attributes claims. The lead agent decomposes queries and describes each subtask (objective, output format, tools, and boundaries) to the subagents.
- Memory
- The lead agent saves its plan to Memory so the plan is retained if the 200,000-token context window is truncated.
- Evals
- On an internal research eval, a multi-agent system with Claude Opus 4 as the lead agent and Claude Sonnet 4 subagents outperformed single-agent Claude Opus 4 by 90.2%. The authors also iterated on prompts by running simulations in the Anthropic Console and watching agents work step-by-step.
- Observability
- not stated
- Monthly cost
- not stated
- Observability tools
- not stated
- Other tools
- Web searchGoogle Workspace integrationsAnthropic Console simulations
- Source
- https://www.anthropic.com/research/multi-agent-research-system
Jeremy Hadfield, Barry Zhang, Kenneth Lien, Florian Scholz, Jeremy Fox, and Daniel Ford · Jun 12, 2025