DLawBench: Evaluating LLMs Through Multi-Turn Legal Consultation

# DLawBench: Evaluating LLMs Through Multi-Turn Legal Consultation
Source: [https://arxiv.org/abs/2606.13931](https://arxiv.org/abs/2606.13931)
Authors:[Li Zhang](https://arxiv.org/search/cs?searchtype=author&query=Zhang,+L),[Yuzhen Shi](https://arxiv.org/search/cs?searchtype=author&query=Shi,+Y),[Yiran Hu](https://arxiv.org/search/cs?searchtype=author&query=Hu,+Y),[Jingwen Zhang](https://arxiv.org/search/cs?searchtype=author&query=Zhang,+J),[Wenbo Lv](https://arxiv.org/search/cs?searchtype=author&query=Lv,+W),[Yubo Ma](https://arxiv.org/search/cs?searchtype=author&query=Ma,+Y),[Wei Wang](https://arxiv.org/search/cs?searchtype=author&query=Wang,+W),[Rongyao Shi](https://arxiv.org/search/cs?searchtype=author&query=Shi,+R),[Yuanyang Qiu](https://arxiv.org/search/cs?searchtype=author&query=Qiu,+Y),[Xinran Xu](https://arxiv.org/search/cs?searchtype=author&query=Xu,+X),[Yuemeng Qi](https://arxiv.org/search/cs?searchtype=author&query=Qi,+Y),[Linlin Miao](https://arxiv.org/search/cs?searchtype=author&query=Miao,+L),[Jaromir Savelka](https://arxiv.org/search/cs?searchtype=author&query=Savelka,+J),[Yun Liu](https://arxiv.org/search/cs?searchtype=author&query=Liu,+Y),[Kevin Ashley](https://arxiv.org/search/cs?searchtype=author&query=Ashley,+K),[Bing Zhao](https://arxiv.org/search/cs?searchtype=author&query=Zhao,+B),[Hu Wei](https://arxiv.org/search/cs?searchtype=author&query=Wei,+H),[Lin Qu](https://arxiv.org/search/cs?searchtype=author&query=Qu,+L)

[View PDF](https://arxiv.org/pdf/2606.13931)

> Abstract:Lawyer\-client consultation is a critical starting point for legal services\. Effective legal assistance hinges on eliciting sufficient and truthful information from clients in order to devise strategies that best protect their interests\. This task requires Large Language Models \(LLMs\) not only to perform robust legal reasoning, but also to strategically elicit material facts through multi\-turn interactions and effectively guide clients with diverse personalities\. Yet existing legal benchmarks overlook this interactive capability\. To fill this gap, we introduce DLawBench, a diagnostic benchmark for real\-world legal consultation\. Drawing on realistic client behavior, we characterize lawyer\-client interactions into four types: Cooperative, Dependent, Withdrawn, and Adversarial\. Using dialogues grounded in real cases, DLawBench evaluates whether LLMs can effectively conduct legal consultation under realistic conditions\. DLawBench comprises 461 cases from Chinese and U\.S\. law, 5,532 paired fact entries, 3,411 inquiry rubrics, and 3,348 issue\-resolution rubrics, and evaluates 26 representative LLMs\. Systematic experiments show substantial headroom: the best\-performing model, GPT\-5\.5, achieves only 0\.562 on consultation\-grounded legal reasoning\. More importantly, DLawBench exposes both sycophancy in legal consultation and a paradox: models perform worse when clients need guidance most\.

## Submission history

From: Li Zhang \[[view email](https://arxiv.org/show-email/6d8a5ff3/2606.13931)\] **\[v1\]**Thu, 11 Jun 2026 21:50:44 UTC \(2,842 KB\)