J1: Incentivizing Thinking in LLM-as-a-Judge via Reinforcement Learning

• URL: http://arxiv.org/abs/2505.10320v3
• Date: Mon, 13 Oct 2025 11:06:51 GMT
• Title: J1: Incentivizing Thinking in LLM-as-a-Judge via Reinforcement Learning
• Authors: Chenxi Whitehouse, Tianlu Wang, Ping Yu, Xian Li, Jason Weston, Ilia Kulikov, Swarnadeep Saha,
• Abstract summary: We introduce J1, a reinforcement learning framework for teaching LLM judges to think before making decisions.<n>Our core contribution lies in converting all judgment tasks for non-verifiable and verifiable prompts into a unified format with verifiable rewards.<n>We then use RL to train thinking-judges at scales of 8B, 32B, and 70B and show that they obtain state-of-the-art performance.
• Score: 54.85131761693927
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: The progress of AI is bottlenecked by the quality of evaluation, making powerful LLM-as-a-Judge models a core solution. The efficacy of these judges depends on their chain-of-thought reasoning, creating a critical need for methods that can effectively optimize this reasoning process. In this work, we introduce J1, a reinforcement learning framework for teaching LLM judges to think before making decisions. Our core contribution lies in converting all judgment tasks for non-verifiable and verifiable prompts into a unified format with verifiable rewards, enabling direct optimization of evaluation quality while mitigating positional bias. We then use RL to train thinking-judges at scales of 8B, 32B, and 70B and show that they obtain state-of-the-art performance across multiple benchmarks. In particular, J1-Qwen-32B, our multitasked pointwise and pairwise judge also outperforms o1-mini, o3, and a much larger 671B DeepSeek-R1 on some benchmarks, while only training on synthetic data. Through comprehensive ablations of pairwise, pointwise, and multitask J1 variants, we demonstrate the effectiveness of our approach across seed prompts, reward strategies, and training recipes. Qualitative analysis reveals that J1 develops systematic evaluation strategies, including dynamic criteria generation, reference answer creation, iterative self-correction of initial assessments, and feedback generation for low-quality responses.

Related papers

• Incentivizing Agentic Reasoning in LLM Judges via Tool-Integrated Reinforcement Learning [30.906073889018728]
  Large Language Models (LLMs) are widely used as judges to evaluate response quality, providing a scalable alternative to human evaluation.<n>We propose TIR-Judge, an end-to-end RL framework for training LLM judges that integrates a code executor for precise evaluation.
  arXiv  Detail & Related papers  (2025-10-27T06:03:37Z)
• Do Before You Judge: Self-Reference as a Pathway to Better LLM Evaluation [22.409170147732464]
  LLM-as-Judge frameworks are increasingly popular for AI evaluation.<n>Research findings on the relationship between models' generation and judgment abilities remain inconsistent.<n>We propose a self-reference-guided evaluation strategy that leverages a model's own answers as references.
  arXiv  Detail & Related papers  (2025-09-24T08:32:45Z)
• LLaVA-Critic-R1: Your Critic Model is Secretly a Strong Policy Model [99.71684530652942]
  We show that LLaVA-Critic-R1 emerges as a top-performing critic but also as a competitive policy model.<n>Applying self-critique at test time yields an average +13.8% improvement on five representative reasoning tasks.<n>Our results reveal that RL training on critic data can produce a unified model excelling at both evaluation and generation.
  arXiv  Detail & Related papers  (2025-08-31T03:08:02Z)
• Incentivizing LLMs to Self-Verify Their Answers [20.2584779107763]
  Large Language Models (LLMs) have demonstrated remarkable progress in complex reasoning tasks.<n>We propose a framework that incentivizes LLMs to self-verify their own answers.<n>We train our self-verification models based on Qwen2.5-Math-7B and DeepSeek-R1-Distill-Qwen-1.5B.
  arXiv  Detail & Related papers  (2025-06-02T06:54:29Z)
• RAG-Zeval: Towards Robust and Interpretable Evaluation on RAG Responses through End-to-End Rule-Guided Reasoning [64.46921169261852]
  RAG-Zeval is a novel end-to-end framework that formulates faithfulness and correctness evaluation as a rule-guided reasoning task.<n>Our approach trains evaluators with reinforcement learning, facilitating compact models to generate comprehensive and sound assessments.<n>Experiments demonstrate RAG-Zeval's superior performance, achieving the strongest correlation with human judgments.
  arXiv  Detail & Related papers  (2025-05-28T14:55:33Z)
• Think-J: Learning to Think for Generative LLM-as-a-Judge [35.036953471299356]
  We propose Think-J, which improves generative LLM-as-a-Judge by learning how to think.<n>We propose two methods for judgment thinking optimization, based on offline and online RL, respectively.<n> Experimental results showed that our approach can significantly enhance the evaluation capability of generative LLM-Judge.
  arXiv  Detail & Related papers  (2025-05-20T12:19:10Z)
• J4R: Learning to Judge with Equivalent Initial State Group Relative Policy Optimization [69.23273504123941]
  We train judges to be robust to positional biases that arise in more complex evaluation settings.<n>We introduce ReasoningJudgeBench, a benchmark that evaluates judges in diverse reasoning settings not covered by prior work.<n>We train Judge for Reasoning (J4R), a 7B judge trained with EIS-GRPO that outperforms GPT-4o and the next best small judge by 6.7% and 9%.
  arXiv  Detail & Related papers  (2025-05-19T16:50:35Z)
• RM-R1: Reward Modeling as Reasoning [81.50471199906738]
  Reasoning Reward Models (ReasRMs) formulate reward modeling as a reasoning task.<n>We propose a reasoning-oriented training pipeline and train a family of ReasRMs, RM-R1.<n>Our models achieve state-of-the-art performance across three reward model benchmarks on average.
  arXiv  Detail & Related papers  (2025-05-05T06:11:12Z)
• VL-Rethinker: Incentivizing Self-Reflection of Vision-Language Models with Reinforcement Learning [55.97950660659051]
  We aim to enhance the slow-thinking capabilities of vision-language models using reinforcement learning (without relying on distillation)<n>We introduce Forced Rethinking, which appends a rethinking trigger token to the end of rollouts in RL training, explicitly enforcing a self-reflection reasoning step.<n>Our model, VL-Rethinker, advances state-of-the-art scores on MathVista, MathVerse to achieve 80.4%, 63.5% respectively.
  arXiv  Detail & Related papers  (2025-04-10T17:41:56Z)
• EQUATOR: A Deterministic Framework for Evaluating LLM Reasoning with Open-Ended Questions. # v1.0.0-beta [2.1249213103048414]
  We introduce the EQUATOR Evaluator, which combines deterministic scoring with a focus on factual accuracy and robust reasoning assessment.<n>Using a vector database, EQUATOR pairs open-ended questions with human-evaluated answers, enabling more precise and scalable evaluations.<n>Our results demonstrate that this framework significantly outperforms traditional multiple-choice evaluations while maintaining high accuracy standards.
  arXiv  Detail & Related papers  (2024-12-31T03:56:17Z)
• CompassJudger-1: All-in-one Judge Model Helps Model Evaluation and Evolution [74.41064280094064]
  textbfJudger-1 is the first open-source textbfall-in-one judge LLM. CompassJudger-1 is a general-purpose LLM that demonstrates remarkable versatility. textbfJudgerBench is a new benchmark that encompasses various subjective evaluation tasks.
  arXiv  Detail & Related papers  (2024-10-21T17:56:51Z)
• SuperCorrect: Advancing Small LLM Reasoning with Thought Template Distillation and Self-Correction [89.56181323849512]
  SuperCorrect is a novel two-stage framework that uses a large teacher model to supervise and correct both the reasoning and reflection processes of a smaller student model.<n>In the first stage, we extract hierarchical high-level and detailed thought templates from the teacher model to guide the student model in eliciting more fine-grained reasoning thoughts.<n>In the second stage, we introduce cross-model collaborative direct preference optimization (DPO) to enhance the self-correction abilities of the student model.
  arXiv  Detail & Related papers  (2024-10-11T17:25:52Z)
• Training Language Models to Self-Correct via Reinforcement Learning [98.35197671595343]
  Self-correction has been found to be largely ineffective in modern large language models (LLMs) We develop a multi-turn online reinforcement learning approach, SCoRe, that significantly improves an LLM's self-correction ability using entirely self-generated data. We find that SCoRe achieves state-of-the-art self-correction performance, improving the base models' self-correction by 15.6% and 9.1% respectively on MATH and HumanEval.
  arXiv  Detail & Related papers  (2024-09-19T17:16:21Z)
• Model-Based Transfer Learning for Contextual Reinforcement Learning [5.5597941107270215]
  We introduce Model-Based Transfer Learning to solve contextual RL problems.<n>We show theoretically that the method exhibits sublinear regret in the number of training tasks.<n>We experimentally validate our methods using urban traffic and standard continuous control benchmarks.
  arXiv  Detail & Related papers  (2024-08-08T14:46:01Z)
• Self-Taught Evaluators [77.92610887220594]
  We present an approach that aims to im-proves without human annotations, using synthetic training data only. Our Self-Taught Evaluator can improve a strong LLM from 75.4 to 88.3 on RewardBench.
  arXiv  Detail & Related papers  (2024-08-05T17:57:02Z)
• Towards Effective Evaluations and Comparisons for LLM Unlearning Methods [97.2995389188179]
  This paper seeks to refine the evaluation of machine unlearning for large language models.<n>It addresses two key challenges -- the robustness of evaluation metrics and the trade-offs between competing goals.
  arXiv  Detail & Related papers  (2024-06-13T14:41:00Z)
• RewardBench: Evaluating Reward Models for Language Modeling [100.28366840977966]
  We present RewardBench, a benchmark dataset and code-base for evaluation of reward models. The dataset is a collection of prompt-chosen-rejected trios spanning chat, reasoning, and safety. On the RewardBench leaderboard, we evaluate reward models trained with a variety of methods.
  arXiv  Detail & Related papers  (2024-03-20T17:49:54Z)
• Aligning Large Language Models by On-Policy Self-Judgment [49.31895979525054]
  Existing approaches for aligning large language models with human preferences face a trade-off that requires a separate reward model (RM) for on-policy learning. We present a novel alignment framework, SELF-JUDGE, that does on-policy learning and is parameter efficient. We show that the rejecting sampling by itself can improve performance further without an additional evaluator.
  arXiv  Detail & Related papers  (2024-02-17T11:25:26Z)
• Prototypical Self-Explainable Models Without Re-training [5.837536154627278]
  Self-explainable models (SEMs) are trained directly to provide explanations alongside their predictions. Current SEMs require complex architectures and heavily regularized loss functions, thus necessitating specific and costly training. We propose a simple yet efficient universal method called KMEx, which can convert any existing pre-trained model into a prototypical SEM.
  arXiv  Detail & Related papers  (2023-12-13T01:15:00Z)
• Effective training-time stacking for ensembling of deep neural networks [1.2667973028134798]
  A snapshot ensembling collects models in the ensemble along a single training path. Our method improves snapshot ensembling by selecting and weighting ensemble members along the training path. It relies on training-time likelihoods without looking at validation sample errors that standard stacking methods do.
  arXiv  Detail & Related papers  (2022-06-27T17:52:53Z)

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.