Research

Single-Agent LLMs Outperform Multi-Agent Systems on Multi-Hop Reasoning Under Equal Thinking Token Budgets

Single-agent LLMs beat multi-agent orchestration on multi-hop reasoning under equal token budgets, suggesting simpler agent architectures may be more computationally efficient than specialized multi-agent setups.

Monday, April 6, 2026 12:00 PM UTC2 MIN READSOURCE: arXiv CS.CL (Computation & Language)BY sys://pipeline

Research shows single-agent LLMs outperform multi-agent systems on multi-hop reasoning tasks when given equal token budgets. This challenges conventional wisdom in agentic AI design and suggests simpler agent architectures may be more efficient than orchestrating multiple specialized agents. Key finding for engineers building reasoning-heavy AI tools.

Read original at arXiv CS.CL (Computation & Language)

ViLegalNLI: Natural Language Inference for Vietnamese Legal Texts

ViLegalNLI enables natural language inference for Vietnamese legal documents, filling a critical gap in legal AI for low-resource languages.

Research1d ago

Are Tools All We Need? Unveiling the Tool-Use Tax in LLM Agents

Research quantifies the performance overhead of tool integration in LLM agents, revealing whether the efficiency cost of tool-use is a fundamental architectural bottleneck.