Unverified50% confidenceFactExact time
Anthropic reported that a multi-agent system led by Claude Opus 4 with Claude Sonnet 4 as sub-agents scored 90.2% higher than single-agent Opus 4 on its internal research evaluation, but consumed about 15 times the tokens of chat; the result has not yet been independently reproduced.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
11/4/2026
First Seen
Valid until: 2/2/2027
Sources
AI Agent Harness Explained: How 66 Frameworks Are Reshaping Agent Capabilities
youtubeAI Briefing Daily11/3/2026
Related Entities
Related Claims
UnverifiedAnthropic reported that its multi-agent research system outperformed a single-agent baseline by 90.2% in internal research evaluations, but consumed about 15 times as many tokens as a standard conversation.65% similarUnverified多项研究表明，多智能体系统中的错误传播率可达单Agent的3-5倍60% similarUnverifiedThe Opus model has a higher per-token cost and consumption than Sonnet, but responds more slowly.58% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1135175API
curl https://kongchang.com/api/v1/knowledge/claims/1135175MCP
get_claim(id=1135175)