The AI legal research startup Descrybe has recently launched DescrybeLM, a new legal reasoning product. The company claims this tool surpasses existing general-purpose AI models, such as ChatGPT, Claude, and Gemini, in a legal reasoning benchmark. A significant feature of this rollout is Descrybe’s publication of the methodology and scoring data, an invitation for external review and verification. The details can be accessed through Descrybe’s white paper.
According to LawNext, the DescrybeLM was rigorously tested against ChatGPT 5.2, Claude Opus 4.5, and Gemini 3 Pro using 200 questions from the National Conference of Bar Examiners’ MBE Complete Practice Exam. DescrybeLM achieved a perfect score by answering all questions correctly. In contrast, the general-purpose AI models fell short, with accuracy rates between 88.5% and 93.5%.
A notable aspect of DescrybeLM’s performance was its near-perfect rubric-scored reasoning quality. The tool scored 99.70%, which measures the systems’ ability to correctly identify and apply legal rules to given facts. This contrasts with the general-purpose AI scores, where discrepancies commonly led to erroneous applications of legal standards, often with an unwarranted confidence.
The company’s co-founder and CEO, Kara Peterson, emphasized the importance of transparency in vendor-produced benchmarks, stating, “Legal professionals deserve to make tool decisions based on real evidence, which can be hard to find. So, we tested ourselves.” The firm has prompted other legal AI vendors to replicate the benchmark and methodology. For further insight, the company has provided a detailed video demonstration and invites legal professionals to engage with their comprehensive documentation.
Despite its apparent superiority, the study acknowledges potential limitations. It indicates the hypothetical risk that the questions employed may have been included in the training data of all evaluated AI systems. Furthermore, Descrybe emphasized that the evaluation was solely focused on multiple-choice legal reasoning and did not cover other legal practices.
The deployment of DescrybeLM signals an evolving landscape in legal AI, one that prioritizes purpose-built tools over general-purpose ones to meet the complexities of legal reasoning and provide more reliable, authoritative analyses.