AI Built for Law Outperforms ChatGPT, Claude, and Gemini on Legal Reasoning Benchmark
DescrybeLM, an AI built specifically for legal reasoning, outperformed ChatGPT, Claude, and Gemini on a 200-question bar exam benchmark.Summary
Descrybe has launched DescrybeLM, an AI system designed specifically for legal reasoning, and published a white paper demonstrating its superior performance compared to general-purpose AI models like ChatGPT 5.2, Claude Opus 4.5, and Gemini 3 Pro. The benchmark involved 200 multistate bar exam questions, evaluating both answer accuracy and the soundness of the legal reasoning. DescrybeLM correctly answered all questions, while the other models missed between 13 and 23. A key finding was that when the general-purpose models were incorrect, they often presented their errors with a high degree of confidence, potentially creating a significant verification burden for legal professionals. The company emphasizes that DescrybeLM was built on a curated legal corpus of over 100 million structured records, distinguishing it from AI tools adapted for legal use. Experts, like Ken Friedman, highlight the substantial difference in performance, categorizing DescrybeLM as a fundamentally different type of tool. The full white paper details the methodology and results, inviting independent replication.
(Source：Ein Presswire)
