Reasoning Jury: Multi-Model Consensus for Evaluating Reasoning Traces

2026-08-12

AI Summary

This work addresses the limitations of existing single-model evaluators, which struggle to accurately identify flaws in long reasoning traces, and the restricted accessibility of state-of-the-art closed-source models for online assessment. To overcome these challenges, the paper introduces Reasoning Jury, a novel framework that employs a panel of open-source large language models—such as gpt-oss-120b—as jurors. Guided by a moderator, these models engage in iterative rounds of critical discussion, opinion revision, and consensus formation to collaboratively judge both the presence and severity of reasoning errors. Without relying on closed-source models, this approach significantly improves identification accuracy and interpretability, outperforming advanced proprietary systems like Opus-4.6, Sonnet-4.6, and Gemini-3.1-Pro while incurring only 8%–15% of their evaluation costs and enabling fine-grained failure mode analysis.

Abstract

Improving reasoning LLMs requires the ability to judge the quality of long reasoning traces for effective reasoning data curation, strong training signals during reinforcement learning, and an in-depth understanding of reasoning behaviors during model performance evaluation. Additionally, surfacing reasoning mistakes that the model makes would enable improving the model's performance at runtime through providing feedback. Due to the difficulty of this complex task on long reasoning traces, single-model judges (even frontier models) do not do well at identifying reasoning defects. Additionally, leveraging frontier models during online training of reasoning LLMs is generally prohibited due to guardrails in terms of use. In this work, we introduce Reasoning Jury, a system that replaces the single judge with a jury of LLMs and a moderated consensus mechanism, to improve the fidelity of judgments for identifying reasoning defects. In reasoning jury, defects of a reasoning trace and their severity are surfaced through a deliberation where a moderator conducts a discussion amongst the jury where the jurors critique each other's judgments and get to modify their initial votes. The moderator derives a consensus through deliberation amongst jurors or consolidation of judgements. We show that Reasoning Jury with a jury of open-weight models (e.g., gpt-oss-120b) is able to significantly outperform frontier models (opus-4.6, sonnet-4.6, and gemini-3.1-pro) at correctly identifying reasoning defects. Besides accuracy performance improvements, the aggregated cost of the jury (initial verdicts, deliberations, consolidation, etc.) is a fraction (8 to 15%) of the cost of running frontier models in LLM-as-a-judge setup. We also show how these judgements can be leveraged to understand failure modes of reasoning LLMs on benchmarks, which allows much deeper understanding of a model's performance.

Problem

• reasoning evaluation
• reasoning defects
• LLM reasoning
• long reasoning traces
• model judgment

Innovation

• Reasoning Jury
• multi-model consensus
• reasoning trace evaluation
• LLM-as-a-judge
• open-weight models