MARS: toward more efficient multi-agent collaboration for LLM reasoning

2025/09/24 by Xiao Wang, Wang, Xiao, Jia Wang +8 · 2 citations
Computer Science · #Artificial Intelligence (cs.AI) #Computation and Language (cs.CL) #FOS: Computer and information sciences #Natural Language Processing Techniques #Semantic Web and Ontologies #Service-Oriented Architecture and Web Services

paper · pdf · doi:10.48550/arxiv.2509.20502

openalex publication_date 2025/09/24 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28

Abstract

Large language models (LLMs) have achieved impressive results in natural language understanding, yet their reasoning capabilities remain limited when operating as single agents. Multi-Agent Debate (MAD) has been proposed to address this limitation by enabling collaborative reasoning among multiple models in a round-table debate manner. While effective, MAD introduces substantial computational overhead due to the number of agents involved and the frequent communication required. In this paper, we propose MARS (Multi-Agent Review System), a role-based collaboration framework inspired by the review process. In MARS, an author agent generates an initial solution, reviewer agents provide decisions and comments independently, and a meta-reviewer integrates the feedback to make the final decision and guide further revision. This design enhances reasoning quality while avoiding costly reviewer-to-reviewer interactions, thereby controlling token consumption and inference time. We compared MARS with both MAD and other state-of-the-art reasoning strategies across multiple benchmarks. Extensive experiments with different LLMs show that MARS matches the accuracy of MAD while reducing both token usage and inference time by approximately 50%. Code is available at https://github.com/xwang97/MARS.

Citations

• LegalReasoner: Step-wised Verification-Correction for Legal Judgment Reasoning
• Enhancing Logical Reasoning in Language Models via Symbolically-Guided Monte Carlo Process Supervision
• ReVISE: Learning to Refine at Test-Time via Intrinsic Self-Verification
• Scoring Verifiers: Evaluating Synthetic Verification for Code and Reasoning
• MM-Verify: Enhancing Multimodal Reasoning with Chain-of-Thought Verification
• AgentRefine: Enhancing Agent Generalization through Refinement Tuning
• Multi-modal Agent Tuning: Building a VLM-Driven Agent for Efficient Tool Usage
• GSM-Symbolic: Understanding the Limitations of Mathematical Reasoning in Large Language Models
• A Peek into Token Bias: Large Language Models Are Not Yet Genuine Reasoners
• Large Language Models: A Survey
• Mixtral of Experts
• Should we be going MAD? A Look at Multi-Agent Debate Strategies for LLMs
• GPQA: A Graduate-Level Google-Proof Q&A Benchmark
• Can LLMs Express Their Uncertainty? An Empirical Evaluation of Confidence Elicitation in LLMs
• Encouraging Divergent Thinking in Large Language Models through Multi-Agent Debate
• Improving Factuality and Reasoning in Language Models through Multiagent Debate
• Tree of Thoughts: Deliberate Problem Solving with Large Language Models
• Reflexion: Language Agents with Verbal Reinforcement Learning
• Large Language Models Can Self-Improve
• Self-Consistency Improves Chain of Thought Reasoning in Language Models
• BNAI, NO-TOKEN, and MIND-UNITY: Pillars of a Systemic Revolution in Artificial Intelligence
• Training Verifiers to Solve Math Word Problems
• Measuring Massive Multitask Language Understanding
• Language Models are Few-Shot Learners
• Deep Residual Learning for Image Recognition

Cited by

• DSCD-Nav: Dual-Stance Cooperative Debate for Object Navigation
• Persona-based Multi-Agent Collaboration for Brainstorming

Related

• Debate Only When Necessary: Adaptive Multiagent Collaboration for Efficient LLM Reasoning
• Towards Reasoning in Large Language Models via Multi-Agent Peer Review Collaboration
• Unleashing Diverse Thinking Modes in LLMs through Multi-Agent Collaboration
• Should we be going MAD? A Look at Multi-Agent Debate Strategies for LLMs
• iMAD: Intelligent Multi-Agent Debate for Efficient and Accurate LLM Inference