# The Cost of Consensus: Isolated Self-Correction Prevails Over Unguided Homogeneous Multi-Agent Debate

arXiv: https://arxiv.org/abs/2605.00914 · HTML: https://mlsanity.com/paper/2605.00914/

## Key facts

- Cited by 2 papers in the map (data through 14 Oct 2026).
- Placed in Social simulation and multi-agent debate, one of 16 directions in LLM agents and tool use.
- Builds on 14 of its 25 references that are in the map.
- First posted on arXiv 29 Apr 2026.

## Placed in

- LLM agents and tool use / [Social simulation and multi-agent debate](https://mlsanity.com/agents/social-simulation-and-multi-agent-debate/)

## Builds on (most cited first, 14 of 14)

- [Scaling Laws for Neural Language Models](https://mlsanity.com/paper/2001.08361/) (2020)
- [Efficient Memory Management for Large Language Model Serving with PagedAttention](https://mlsanity.com/paper/2309.06180/) (2023)
- [MMLU-Pro: A More Robust and Challenging Multi-Task Language Understanding Benchmark](https://mlsanity.com/paper/2406.01574/) (2024)
- [Improving Factuality and Reasoning in Language Models through Multiagent Debate](https://mlsanity.com/paper/2305.14325/) (2023)
- [Towards Understanding Sycophancy in Language Models](https://mlsanity.com/paper/2310.13548/) (2023)
- [Encouraging Divergent Thinking in Large Language Models through Multi-Agent Debate](https://mlsanity.com/paper/2305.19118/) (2023)
- [Discovering Language Model Behaviors with Model-Written Evaluations](https://mlsanity.com/paper/2212.09251/) (2022)
- [Large Language Models Sensitivity to The Order of Options in Multiple-Choice Questions](https://mlsanity.com/paper/2308.11483/) (2023)
- [MultiAgentBench: Evaluating the Collaboration and Competition of LLM agents](https://mlsanity.com/paper/2503.01935/) (2025)
- [Talk Isn't Always Cheap: Understanding Failure Modes in Multi-Agent Debate](https://mlsanity.com/paper/2509.05396/) (2025)
- [Optimizing Model Selection for Compound AI Systems](https://mlsanity.com/paper/2502.14815/) (2025)
- [Peacemaker or Troublemaker: How Sycophancy Shapes Multi-Agent Debate](https://mlsanity.com/paper/2509.23055/) (2025)
- [S2-MAD: Breaking the Token Barrier to Enhance Multi-Agent Debate Efficiency](https://mlsanity.com/paper/2502.04790/) (2025)
- [Power and Limitations of Aggregation in Compound AI Systems](https://mlsanity.com/paper/2602.21556/) (2026)

## Built on by (most cited first, 2 of 2)

- [Engineering Reliable Coding Agents: Evaluating and Operating the System Around the Model](https://mlsanity.com/paper/2608.13867/) (2026)
- [CollabFlow: Recursive Self-Improvement of Agent Collaboration](https://mlsanity.com/paper/2609.38662/) (2026)

## Abstract (arXiv)

Multi-agent debate, where teams of LLMs iteratively exchange rationales and vote on answers, is widely deployed under the assumption that peer review filters hallucinations. Yet the failure dynamics of homogeneous debate remain poorly understood, therefore we report findings from a controlled empirical study of teams of N=10 homogeneous agents (Qwen2.5-7B, Llama-3.1-8B, Ministral-3-8B) across R=3 debate rounds on two high-difficulty benchmarks (GSM-Hard and MMLU-Hard). We compare peer debate against isolated self-correction and a stochastic noise control that injects rationales from unrelated problems. We decompose debate failure into three model-dependent pathways: sycophantic conformity, where agents uncritically adopt majority answers (modal adoption up to 85.5%); contextual fragility, where peer rationales destabilize previously correct reasoning (vulnerability rate up to 70.0%); and consensus collapse, where plurality voting discards correct answers already present in the generation pool (oracle gap up to 32.3 percentage points). Ablations over communication density (K ∈ 2,4,9) and sampling temperature (T ∈ 0.4, 0.7) show that conformity reaches high levels at minimal peer exposure (K=2) and intensifies with greater initial diversity. Across all configurations, debate consumes 2.1-3.4× more tokens (up to 28,631 tokens per problem) than self-correction for equal or lower accuracy. Our results indicate that, within the 7-8B parameter class, homogeneous teams without structured roles do not benefit from unguided peer exchange, and that isolated self-correction consistently offers a more favorable cost-accuracy tradeoff.