mlsanityData to 14 Oct 2026

Where Do LLM Ensemble Gains Come From? Decomposing Candidate Value, Diversity, and Aggregation

arXiv 2610.18786 ↗First posted 13 Oct 2026cs.CL cs.AIcited by 0

Not on the map. Why?

Abstract (arXiv)

LLM ensemble gains are often measured against a single baseline, but the value attributed to additional candidates can change with the comparison used. We study an LLM aggregator under four matched conditions: Direct (no candidates), Single-slot (one candidate in a multi-slot interface), Duplicate (one candidate repeated), and Multi-candidate (separately generated candidates). Under either non-diverse control, overall candidate gain (CIV) splits algebraically into an additional-candidate contrast (DSV) and a candidate-context contrast (CCV); the split, but not CIV, depends on the chosen control. On MultiHopRAG model diversity (k=2), the repetition gap Rep=-8.33 pp [-13.25, -3.75] quantifies the difference between gains measured against Duplicate and Single-slot. On a matched FRAMES stronger-panel extension (n=200, k=3, aggregator output limit 4096), paired Delta Rep=+9.67 pp [+6.00, +13.50]; within panels, Rep changes sign (original -6.33; stronger +3.33). In FRAMES evidence aggregation, candidate-context contrasts remain strongly negative under both non-diverse controls. Our results show why ensemble evaluations should state the reference control before attributing gains to additional candidates.

Key facts. No citing papers in the map yet (data through 14 Oct 2026). Builds on 19 of its 32 references that are in the map. First posted on arXiv 13 Oct 2026.

Builds on

19 of 32 references.

References · 19

  1. 2023Judging LLM-as-a-Judge with MT-Bench and Chatbot Arenacited by 8,626
  2. 2023Lost in the Middle: How Language Models Use Long Contextscited by 3,455
  3. 2023Improving Factuality and Reasoning in Language Models through Multiagent Debatecited by 1,784
  4. 2023LLM-Blender: Ensembling Large Language Models with Pairwise Ranking and Generative Fusioncited by 564
  5. 2024Mixture-of-Agents Enhances Large Language Model Capabilitiescited by 438
  6. 2024MultiHop-RAG: Benchmarking Retrieval-Augmented Generation for Multi-Hop Queriescited by 257
  7. 2023Universal Self-Consistency for Large Language Model Generationcited by 203
  8. 2024Fact, Fetch, and Reason: A Unified Evaluation of Retrieval-Augmented Generationcited by 162
  9. 2023Should we be going MAD? A Look at Multi-Agent Debate Strategies for LLMscited by 114
  10. 2025Voting or Consensus? Decision-Making in Multi-Agent Debatecited by 59
  11. 2025Rethinking Mixture-of-Agents: Is Mixing Different Large Language Models Beneficial?cited by 54
  12. 2026When Does Combining Language Models Help? A Co-Failure Ceiling on Routing, Voting, and Mixture-of-Agents Across 67 Frontier Modelscited by 13
7 more
  1. 2026Are Diversity Metrics Measuring Diversity? A Capability-Controlled Audit of Majority-Vote Gain in LLM Ensemblescited by 9
  2. 2026Rational Synthesizers or Heuristic Followers? Analyzing LLMs in RAG-based Question-Answeringcited by 5
  3. 2026When Agents Disagree: The Selection Bottleneck in Multi-Agent LLM Pipelinescited by 5
  4. 2026Quantifying Diversity of Thought: A Predictive Law of Weighted LLM Ensemble Liftcited by 5
  5. 2026Beyond Consensus: Trace-Level Synthesis in Mixture of Agentscited by 4
  6. 2026LLM-as-a-Judge Scores Are Unreliable Optimization Signals in Closed-Loop Table Recognitioncited by 3
  7. 2026When LLMs Sit Above Diagnostic Tools: Unrealized Complementarity in Industrial Fault Diagnosiscited by 1

Built on by

Not cited yet.