Unverified50% confidenceFactExact time
A Stanford preprint paper, "Single-Agent LLMs Outperform Multi-Agent Systems on Multi-Hop Reasoning Under Equal Thinking Token Budgets," held the reasoning token budget constant across three model families and found that the single agent consistently matched or outperformed the multi-agent setup, explaining the phenomenon with the data processing inequality.
1
Sources
50%
Confidence
Long-term
Relevance
10/14/2026
First Seen
Sources
Related Entities
Related Claims
Unverified多智能体系统通过多个LLM实例各自承担独立角色，彼此质疑和交叉验证形成自我纠错机制，在复杂推理任务上往往优于单一模型独立工作70% similarUnverifiedSelf-consistency typically samples 3-5 reasoning paths and takes a majority vote; the cost is several times that of a single call, and it is suitable for math and logic problems with a unique correct answer.63% similarUnverifiedSub-agents independently select models based on their own assessment of subtask complexity rather than inheriting the model setting of the main session, which causes multiple models that the user never manually selected to appear in usage reports.63% similarUnverifiedThe ReAct-style autonomous decision-making loop of agentic retrieval enables it to handle questions that require multi-hop reasoning.63% similarUnverifiedWhen the reasoning model, personality settings, and long-term memory are independent of one another, rapid iteration of the underlying model does not force the project to start over from scratch.63% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1025946API
curl https://kongchang.com/api/v1/knowledge/claims/1025946MCP
get_claim(id=1025946)