mlsanityData to 14 Oct 2026

Attention Tax, Handoff Tax: A Stylised Model of When Multi-Agent LLM Systems Help

arXiv 2610.06069 ↗First posted 5 Oct 2026cs.MA cs.AI cs.CLcited by 0

On the map: LLM agents and tool use → General LLM agent frameworks and benchmarks also fits: Agent memory systems

Abstract (arXiv)

Recent work on multi-agent LLM systems reaches sharply different conclusions: some results show that a single agent with the same information and compute should dominate a delegated system, others that multi-agent gains grow with task depth. We argue that much of the disagreement comes from modelling different bottlenecks, and introduce a stylised reliability model built around two trade-offs. Decomposition reduces the burden of long contexts but incurs a handoff tax when information is compressed or transferred between agents. Redundancy gains from multiple samples, but its benefit depends on how much their failures are shared. With reasoning budget, verification, and task structure added, the model yields two crossover conditions: decomposition becomes preferable once the attention cost avoided by resetting context exceeds the handoff cost, and parallel sampling at equal budget is eventually preferable when its shared-failure floor lies below the error floor of one agent thinking longer. We connect these regimes to recent theoretical and empirical results. On a ledger-reconciliation task we measure the context-degradation curve and the handoff tax from single-agent and handoff runs alone. From these the model places the crossover at depth 10 and predicts decomposition to win at depths 20, 50, and 100. It does, on step-level and final-balance accuracy, and the decomposed system's success, which the prediction never sees, lands within 9 percentage points of the predicted rate at every depth.

Key facts. No citing papers in the map yet (data through 14 Oct 2026). Placed in General LLM agent frameworks and benchmarks, one of 16 directions in LLM agents and tool use. Builds on 33 of its 39 references that are in the map. First posted on arXiv 5 Oct 2026.

Builds on

33 of 39 references.

Closest ancestors · 7

Same direction, newest first

  1. 2026Single-Agent LLMs Outperform Multi-Agent Systems on Multi-Hop Reasoning Under Equal Thinking Token Budgets2 Apr
  2. 2026On the Reliability Limits of LLM-Based Multi-Agent Planning27 Mar
  3. 2026When Agents Disagree: The Selection Bottleneck in Multi-Agent LLM Pipelines20 Mar
  4. 2026Understanding Agent Scaling in LLM-Based Multi-Agent Systems via Diversity3 Feb
  5. 2026Phase Transition for Budgeted Multi-Agent Synergy24 Jan
  6. 2025Towards a Science of Scaling Agent Systems9 Dec
  7. 2025On the Importance of Task Complexity in Evaluating LLM-Based Multi-Agent Systems5 Oct

Other references · 26

  1. 2022Self-Consistency Improves Chain of Thought Reasoning in Language Modelscited by 5,708
  2. 2023Lost in the Middle: How Language Models Use Long Contextscited by 3,455
  3. 2024Scaling LLM Test-Time Compute Optimally can be More Effective than Scaling Model Parameterscited by 1,986
  4. 2024RULER: What's the Real Context Size of Your Long-Context Language Models?cited by 1,346
  5. 2023Large Language Models Cannot Self-Correct Reasoning Yetcited by 1,017
  6. 2025Why Do Multi-Agent LLM Systems Fail?cited by 604
  7. 2025LLMs Get Lost In Multi-Turn Conversationcited by 429
  8. 2025Scaling Long-Horizon LLM Agent via Context-Foldingcited by 148
  9. 2025ACON: Optimizing Context Compression for Long-horizon LLM Agentscited by 136
  10. 2025Context Length Alone Hurts LLM Performance Despite Perfect Retrievalcited by 127
  11. 2025NoLiMa: Long-Context Evaluation Beyond Literal Matchingcited by 126
  12. 2025Debate or Vote: Which Yields Better Decisions in Multi-Agent Large Language Models?cited by 81
14 more
  1. 2025The Illusion of Diminishing Returns: Measuring Long Horizon Execution in LLMscited by 64
  2. 2026Harness-Bench: Measuring Harness Effects across Models in Realistic Agent Workflowscited by 57
  3. 2026Stop Comparing LLM Agents Without Disclosing the Harnesscited by 34
  4. 2026Rethinking the Value of Multi-Agent Workflow: A Strong Single Agent Baselinecited by 22
  5. 2025Solving a Million-Step LLM Task with Zero Errorscited by 19
  6. 2026Auditing Multi-Agent LLM Reasoning Trees Outperforms Majority Vote and LLM-as-Judgecited by 18
  7. 2025When Does Divide and Conquer Work for Long Context LLM? A Noise Decomposition Frameworkcited by 9
  8. 2026The Illusion of Multi-Agent Advantagecited by 9
  9. 2026The Handoff Tax: Continuing Non-Native Trajectories in LLM Agentscited by 6
  10. 2026The Six Sigma Agent: Achieving Enterprise-Grade Reliability in LLM Systems Through Consensus-Driven Decomposed Executioncited by 4
  11. 2026Breaking the Martingale Curse: Multi-Agent Debate via Asymmetric Cognitive Potential Energycited by 3
  12. 2026Diagnosing and Mitigating Context Rot in Long-horizon Searchcited by 3
  13. 2026When Do Multi-Agent Systems Outperform? Analysing the Learning Efficiency of Agentic Systemscited by 2
  14. 2026On the Reliability of Networks of AI Agents: Density Evolution, Stopping Sets, and Architecture Optimizationcited by 1

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2026Beyond Token Savings: A Systematic Study of Context Compression in LLM Agents
  2. 2026Multi-agent Scaling Across Disjunctive and Compensatory Tasks
  3. 2026The Collaboration Tax: How Much LLM Multi-Agent Systems Pay to Coordinate
  4. 2026The Epistemics of Agent Memory: Measuring, and Governing, the Consolidation Decision in Long-Horizon LLM Agents
  5. 2026Bilevel Coordinated Reflection: A Game-Theoretic Approach to Multi-Agent LLM Systems
  6. 2025United Minds or Isolated Agents? Exploring Coordination of LLMs under Cognitive Load Theory
  7. 2025Self-Resource Allocation in Multi-Agent LLM Systems
  8. 2025How Memory Management Impacts LLM Agents: An Empirical Study of Experience-Following Behavior