LLMs as a Jury: Cross-Model Consensus Can Outperform Process Reward Models for LLM Reasoning

• URL: http://arxiv.org/abs/2607.10139v2
• Date: Mon, 20 Jul 2026 01:48:49 GMT
• Title: LLMs as a Jury: Cross-Model Consensus Can Outperform Process Reward Models for LLM Reasoning
• Abstract summary: We study cross-model consensus, the degree to which independently trained models, each solving the problem once, agree on a final answer.<n>We treat the panel as an LLM-jury, in which the structure of agreement, not any model's score of another, is the verification signal.<n>We make this precise with a parameter-free law, derived in closed form, that predicts consensus accuracy from three measured panel statistics to a mean absolute error of $0.03$.
• Score: 6.241883798820154
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Selecting the correct answer from a pool of candidate reasoning chains is the engine of test-time scaling, yet the standard selectors each carry a cost: self-consistency inherits the errors of the single model it resamples, and trained reward models need labeled data and transfer poorly off-distribution. We study a third signal, free at inference time: cross-model consensus, the degree to which independently trained models, each solving the problem once, agree on a final answer. We treat the panel as an LLM-jury, in which the structure of agreement, not any model's score of another, is the verification signal. Across seven benchmarks it selects correct answers better than self-consistency and far better than a model scoring its own candidates: on competition math it closes the entire gap to an oracle selector, while self-scoring closes almost none. The mechanism is error decorrelation: independently trained models err differently, so their wrong answers scatter while the correct one accumulates agreement. We make this precise with a parameter-free law, derived in closed form, that predicts consensus accuracy from three measured panel statistics to a mean absolute error of $0.03$ and exposes the method's ceiling: a shared-error floor where models share a misconception, near zero on math but non-trivial on science. Against four trained verifiers spanning discriminative, outcome, and generative reward models, the free LLM-jury matches the strongest inside their math training domain and is the top selector outside it. Cross-model consensus is thus a verifier we can characterize in advance: a law that says when to trust it, and a floor that marks where it cannot.

Related papers

• Check The Scoreboard: An Analysis of Scoring Schemes on Multiple-Choice Evaluation [52.58737865652009]
  Multiple-choice question answering (MCQA) benchmarks in NLP use number-right scoring (accuracy)<n>In educational testing, the scoring scheme, the combination of the response mode models follow and the rule for grading responses, is a key design choice that dictates which abilities to reward.<n>We examine how alternatives to number right change what MCQA measures with six education-inspired schemes that assess abilities beyond accuracy.
  arXiv  Detail & Related papers  (2026-08-30T16:35:22Z)
• Measuring Reward Hacking and Reasoning-Answer Decoupling Under Position-Confounded Optimization [1.1316247605466565]
  We train language models with GRPO on math problems where the correct answer is always option A, then evaluate on an unseen test set with unbiased answer positions.<n>Across Qwen2.5, Llama 3.x and Gemma-3 models, biased training often drives option-A rates above 0.90 in smaller models.<n>We track reasoning-answer decoupling: capable models generate reasoning that reaches the correct numeric answer while still selecting A.
  arXiv  Detail & Related papers  (2026-08-15T23:13:57Z)
• Learning When to Trust via Selective Context Preference Optimization [34.54211638111961]
  Language models increasingly condition their answers on external signals.<n>A single misleading one can turn a correct answer wrong.<n>We recast the problem as selective trust.
  arXiv  Detail & Related papers  (2026-08-06T17:59:58Z)
• When LLMs Agree, Are They Right? Auditing Self-Consistency and Cross-Model Agreement as Confidence Signals [0.0]
  LLM-as-judge is increasingly the default for evaluating AI systems in enterprise pipelines.<n>We show that consistency -- agreement among judges, or a model's own samples -- indicates correctness.<n>We ask when agreement is nonetheless a usable proxy, in a large-scale cross-runner study.
  arXiv  Detail & Related papers  (2026-07-09T02:46:51Z)
• Self-Verified Distillation: Your Language Model Is Secretly Its Own Synthetic Data Pipeline [56.53954182896384]
  We propose Self-Verified Distillation, a simple post-training refinement algorithm for large language models.<n>Self-Verified Distillation generates candidate solutions to unlabeled seed questions.<n>It filters them using prompt-based self-verification, and trains on the resulting self-curated dataset.<n>We find that sampling more candidate generations and using a larger verification budget during training data construction produces higher-quality self-curated data.
  arXiv  Detail & Related papers  (2026-05-20T17:26:10Z)
• Cross-Model Disagreement as a Label-Free Correctness Signal [10.66607150500579]
  Cross-model disagreement is a correctness indicator that can be dropped into existing production systems.<n>No generation from the verifying model is required, and no correctness labels are needed.<n>Results establish cross-model disagreement as a practical, training-free approach to label-free correctness estimation.
  arXiv  Detail & Related papers  (2026-03-26T13:46:22Z)
• Dependence-Aware Label Aggregation for LLM-as-a-Judge via Ising Models [55.94503936470247]
  Large-scale AI evaluation increasingly relies on aggregating binary judgments from $K$ annotators, including judges.<n>Most classical methods assume annotators are conditionally independent given the true label $Yin0,1$, an assumption often violated by LLM judges.<n>We study label aggregation through a hierarchy of dependence-aware models based on Ising graphical models and latent factors.
  arXiv  Detail & Related papers  (2026-01-29T21:26:50Z)
• How Can I Publish My LLM Benchmark Without Giving the True Answers Away? [3.4155322317700585]
  Publishing a large language model (LLM) benchmark on the Internet risks contaminating future LLMs.<n>A common mitigation is to keep the benchmark private and let participants submit their models or predictions to the organizers.<n>We propose a way to publish benchmarks without completely disclosing the ground-truth answers to the questions.
  arXiv  Detail & Related papers  (2025-05-23T16:57:34Z)
• Ranked from Within: Ranking Large Multimodal Models Without Labels [73.96543593298426]
  We show that uncertainty scores derived from softmax distributions provide a robust basis for ranking models across various tasks.<n>This facilitates the ranking of LMMs on unlabeled data, providing a practical approach for selecting models for diverse target domains without requiring manual annotation.
  arXiv  Detail & Related papers  (2024-12-09T13:05:43Z)
• LACIE: Listener-Aware Finetuning for Confidence Calibration in Large Language Models [69.68379406317682]
  We introduce a listener-aware finetuning method (LACIE) to calibrate implicit and explicit confidence markers. We show that LACIE models the listener, considering not only whether an answer is right, but whether it will be accepted by a listener. We find that training with LACIE results in 47% fewer incorrect answers being accepted while maintaining the same level of acceptance for correct answers.
  arXiv  Detail & Related papers  (2024-05-31T17:16:38Z)
• AvgOut: A Simple Output-Probability Measure to Eliminate Dull Responses [97.50616524350123]
  We build dialogue models that are dynamically aware of what utterances or tokens are dull without any feature-engineering. The first model, MinAvgOut, directly maximizes the diversity score through the output distributions of each batch. The second model, Label Fine-Tuning (LFT), prepends to the source sequence a label continuously scaled by the diversity score to control the diversity level. The third model, RL, adopts Reinforcement Learning and treats the diversity score as a reward signal.
  arXiv  Detail & Related papers  (2020-01-15T18:32:06Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.