LLM作为验证器：一种通用验证框架

摘要

扩展预训练、后训练和测试时计算的规模已成为提升大语言模型能力的主要范式。在本工作中，我们将“验证”（即判断解决方案正确性的能力）定义为一条新的扩展维度。为解锁这一能力并展示其有效性，我们提出了LLM-as-a-Verifier（作为验证器的大语言模型），这是一个通用验证框架，能够在无需额外训练的情况下为智能体任务提供细粒度反馈。与标准的大语言模型裁判（通过提示模型对候选方案生成离散分数）不同，LLM-as-a-Verifier通过计算评分词元对数几率分布上的期望值来生成连续分数。这种概率化表达方式使验证能够在多个维度上实现扩展：（1）评分粒度，（2）重复评估，以及（3）标准分解。具体而言，我们证明扩展评分粒度能够更好地区分正面与负面解决方案，从而实现更精确的比较。此外，扩展重复评估和标准分解通过降低方差和复杂度，持续提升验证准确性。我们还提出了一种基于验证器连续分数的成本高效排序算法，用于从候选方案中筛选最佳解决方案。LLM-as-a-Verifier在Terminal-Bench V2（86.5%）、SWE-Bench Verified（78.2%）、RoboRewardBench（87.4%）和MedAgentBench（73.3%）上取得了最先进性能。除验证外，LLM-as-a-Verifier提供的细粒度信号还可作为估算任务进度的代理指标。我们为Claude Code构建了扩展功能，使开发者能够监控和改进自身的智能体系统。最后，研究表明LLM-as-a-Verifier能为强化学习提供密集反馈，在机器人和数学推理基准上提升SAC和GRPO的样本效率。

English

Scaling pre-training, post-training, and test-time compute have become the central paradigms for improving the capabilities of LLMs. In this work, we identify verification, the ability to determine the correctness of a solution, as a new scaling axis. To unlock this and demonstrate its effectiveness, we introduce LLM-as-a-Verifier, a general-purpose verification framework that provides fine-grained feedback for agentic tasks without requiring additional training. Unlike standard LM judges that prompt LLMs to produce discrete scores for candidate solutions, LLM-as-a-Verifier computes the expectation over the distribution of scoring token logits to generate continuous scores. This probabilistic formulation enables verification to scale along multiple dimensions: (1) score granularity, (2) repeated evaluation, and (3) criteria decomposition. In particular, we show that scaling the scoring granularity leads to better separation between positive and negative solutions, resulting in more calibrated comparisons. Moreover, scaling repeated evaluation and criteria decomposition consistently lead to additional gains in verification accuracy through variance and complexity reduction. We further introduce a cost-efficient ranking algorithm for selecting the best solution among candidates using the verifier's continuous scores. LLM-as-a-Verifier achieves state-of-the-art performance on Terminal-Bench V2 (86.5%), SWE-Bench Verified (78.2%), RoboRewardBench (87.4%), and MedAgentBench (73.3%). Beyond verification, the fine-grained signals from LLM-as-a-Verifier can also serve as a proxy for estimating task progress. We build an extension for Claude Code, enabling developers to monitor and improve their own agentic systems. Finally, we show that LLM-as-a-Verifier can provide dense feedback for RL, improving the sample efficiency of SAC and GRPO on robotics and mathematical reasoning benchmarks.