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

arXiv: https://arxiv.org/abs/2604.02460 · HTML: https://mlsanity.com/paper/2604.02460/

## Key facts

- Cited by 30 papers in the map (data through 14 Oct 2026).
- Placed in Reasoning for agents, planning and code, one of 17 directions in LLM reasoning.
- Builds on 20 of its 28 references that are in the map.
- First posted on arXiv 2 Apr 2026.

## Placed in

- LLM reasoning / [Reasoning for agents, planning and code](https://mlsanity.com/lr/reasoning-for-agents-planning-and-code/)
- LLM agents and tool use / [General LLM agent frameworks and benchmarks](https://mlsanity.com/agents/general-llm-agent-frameworks-and-benchmarks/)

## Builds on (most cited first, 20 of 20)

- [Reflexion: Language Agents with Verbal Reinforcement Learning](https://mlsanity.com/paper/2303.11366/) (2023)
- [Lost in the Middle: How Language Models Use Long Contexts](https://mlsanity.com/paper/2307.03172/) (2023)
- [Improving Factuality and Reasoning in Language Models through Multiagent Debate](https://mlsanity.com/paper/2305.14325/) (2023)
- [MuSiQue: Multihop Questions via Single-hop Question Composition](https://mlsanity.com/paper/2108.00573/) (2021)
- [Why Do Multi-Agent LLM Systems Fail?](https://mlsanity.com/paper/2503.13657/) (2025)
- [Token-Budget-Aware LLM Reasoning](https://mlsanity.com/paper/2412.18547/) (2024)
- [Exploring Collaboration Mechanisms for LLM Agents: A Social Psychology View](https://mlsanity.com/paper/2310.02124/) (2023)
- [More Agents Is All You Need](https://mlsanity.com/paper/2402.05120/) (2024)
- [Fact, Fetch, and Reason: A Unified Evaluation of Retrieval-Augmented Generation](https://mlsanity.com/paper/2409.12941/) (2024)
- [Towards a Science of Scaling Agent Systems](https://mlsanity.com/paper/2512.08296/) (2025)
- [Context Length Alone Hurts LLM Performance Despite Perfect Retrieval](https://mlsanity.com/paper/2510.05381/) (2025)
- [SelfBudgeter: Adaptive Token Allocation for Efficient LLM Reasoning](https://mlsanity.com/paper/2505.11274/) (2025)
- [Reasoning in Token Economies: Budget-Aware Evaluation of LLM Reasoning Strategies](https://mlsanity.com/paper/2406.06461/) (2024)
- [MAS-Orchestra: Understanding and Improving Multi-Agent Reasoning Through Holistic Orchestration and Controlled Benchmarks](https://mlsanity.com/paper/2601.14652/) (2026)
- [MedAgentBoard: Benchmarking Multi-Agent Collaboration with Conventional Methods for Diverse Medical Tasks](https://mlsanity.com/paper/2505.12371/) (2025)
- [Single-agent or Multi-agent Systems? Why Not Both?](https://mlsanity.com/paper/2505.18286/) (2025)
- [BudgetThinker: Empowering Budget-aware LLM Reasoning with Control Tokens](https://mlsanity.com/paper/2508.17196/) (2025)
- [A Controlled Study on Long Context Extension and Generalization in LLMs](https://mlsanity.com/paper/2409.12181/) (2024)
- [FocusLLM: Precise Understanding of Long Context by Dynamic Condensing](https://mlsanity.com/paper/2408.11745/) (2024)
- [Single-Agent vs. Multi-Agent LLM Strategies for Automated Student Reflection Assessment](https://mlsanity.com/paper/2504.05716/) (2025)

## Built on by (most cited first, 30 of 30)

- [Recursive Harness Self-Improvement](https://mlsanity.com/paper/2607.15524/) (2026)
- [The Illusion of Multi-Agent Advantage](https://mlsanity.com/paper/2606.13003/) (2026)
- [FutureSim: Replaying World Events to Evaluate Adaptive Agents](https://mlsanity.com/paper/2605.15188/) (2026)
- [When the Tool Decides: LLM Agents Defer Blindly to Graph Neural Network Tools, and Stronger Backbones Defer More](https://mlsanity.com/paper/2606.14476/) (2026)
- [The Interaction Tax: When Communication Erases Diversity in Multi-Agent Teams](https://mlsanity.com/paper/2608.23541/) (2026)
- [Towards Optimal Agentic Architectures for Offensive Security Tasks](https://mlsanity.com/paper/2604.18718/) (2026)
- [Evolve as a Team: Collaborative Self-Evolution for LLM-based Multi-Agent Systems](https://mlsanity.com/paper/2605.29790/) (2026)
- [The Consistency Illusion: How Multi-Agent Debate Hides Reasoning Misalignment](https://mlsanity.com/paper/2606.08457/) (2026)
- [Oracle Gap and Signal Fidelity: A Fixed-Pool Diagnostic for Test-Time Collaboration](https://mlsanity.com/paper/2607.17531/) (2026)
- [From Language Models to World-Acting Systems: Progress and Limits of Agentic AI across Digital, Social, Virtual, and Physical Environments](https://mlsanity.com/paper/2609.04894/) (2026)
- [Agentic Fuzzing: Opportunities and Challenges](https://mlsanity.com/paper/2605.10074/) (2026)
- [One Model, Many Minds: Unlocking Multi-Agent Synergy in a Single Agent via Mixture of Roles](https://mlsanity.com/paper/2608.27338/) (2026)
- [OpenCollab: A Multi-Agent Coding Framework with Programmable Collaboration and Controllable Runtime](https://mlsanity.com/paper/2609.38345/) (2026)
- [Multi-Agent Systems are Mixtures of Experts: Who Becomes an Influencer?](https://mlsanity.com/paper/2605.25929/) (2026)
- [Are Large Language Models Suitable for Graph Computation? Progress and Prospects](https://mlsanity.com/paper/2606.06865/) (2026)
- [Faithful, Not Corrective: Model Capability Governs Message-Format Effects in Multi-Hop Agent Relays](https://mlsanity.com/paper/2607.09678/) (2026)
- [When Do Multi-Agent Systems Help? An Information Bottleneck Perspective](https://mlsanity.com/paper/2607.16133/) (2026)
- [Sample More, Reflect Less: Self-Refine and Reflexion Lose to Repeated Sampling at Equal Token Cost, from 1.5B to 7B](https://mlsanity.com/paper/2607.28576/) (2026)
- [Width, Memory, and Delay: A Resource Accounting for the Limits of Flat Multi-Agent Systems](https://mlsanity.com/paper/2608.00028/) (2026)
- [When Does LLM Orchestration Pay Off? A Controlled Evaluation of Accuracy, Cost, and Task Difficulty](https://mlsanity.com/paper/2608.00685/) (2026)
- [Don't Overthink, Don't Underthink: Toward Adaptive Reasoning in Agentic AI](https://mlsanity.com/paper/2608.26442/) (2026)
- [GVS5H: Zero-Shot Self-Orchestration with Ledger-Based Control for Improved LLM Coding Performance](https://mlsanity.com/paper/2608.26480/) (2026)
- [Perceive to Hypothesize, Verify to Ground: An Agentic Reasoning Framework for Open-World Geo-Localization](https://mlsanity.com/paper/2608.29880/) (2026)
- [What Do CAE Simulation Agents Really Need Beyond a Generic Harness?](https://mlsanity.com/paper/2609.03718/) (2026)
- [The Communication Bottleneck: A Round-Trip Study of Tree-Structured Expression Serialization in Language Models](https://mlsanity.com/paper/2609.21509/) (2026)
- [What Confidence Routing Is Actually Doing: Auditing Routing, Calibration, and Commitment in Multi-Agent Deliberation](https://mlsanity.com/paper/2609.27822/) (2026)
- [CP-Agent: A Harness-Engineered Agent for Crystal Plasticity Simulation Workflows](https://mlsanity.com/paper/2609.31790/) (2026)
- [From Discovery to Decision: Finite-Budget Recoverability in LLM Voting](https://mlsanity.com/paper/2610.01014/) (2026)
- [Attention Tax, Handoff Tax: A Stylised Model of When Multi-Agent LLM Systems Help](https://mlsanity.com/paper/2610.06069/) (2026)
- [Evolve Together, Fall into Place: How Social Scaffolding Shapes Self-Evolving Multi-Agent Systems](https://mlsanity.com/paper/2610.19799/) (2026)

## Abstract (arXiv)

Recent work reports strong performance from multi-agent LLM systems (MAS), but these gains are often confounded by increased test-time computation. When computation is normalized, single-agent systems (SAS) can match or outperform MAS, yet the theoretical basis and evaluation methodology behind this comparison remain unclear. We present an information-theoretic argument, grounded in the Data Processing Inequality, suggesting that under a fixed reasoning-token budget and with perfect context utilization, single-agent systems are more information-efficient. This perspective further predicts that multi-agent systems become competitive when a single agent's effective context utilization is degraded, or when more compute is expended. We test these predictions in a controlled empirical study across three model families (Qwen3, DeepSeek-R1-Distill-Llama, and Gemini 2.5), comparing SAS with multiple MAS architectures under matched budgets. We find that SAS consistently match or outperform MAS on multi-hop reasoning tasks when reasoning tokens are held constant. Beyond aggregate performance, we conduct a detailed diagnostic analysis of system behavior and evaluation methodology. We identify significant artifacts in API-based budget control (particularly in Gemini 2.5) and in standard benchmarks, both of which can inflate apparent gains from MAS. Overall, our results suggest that, for multi-hop reasoning tasks, many reported advantages of multi-agent systems are better explained by unaccounted computation and context effects rather than inherent architectural benefits, and highlight the importance of understanding and explicitly controlling the trade-offs between compute, context, and coordination in agentic systems.