<?xml version="1.0" encoding="UTF-8"?>
<!DOCTYPE root>
<article xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink" xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance" xmlns:ali="http://www.niso.org/schemas/ali/1.0/" article-type="research-article" dtd-version="1.2" xml:lang="en"><front><journal-meta><journal-id journal-id-type="publisher-id">Gaps in Russian Legislation</journal-id><journal-title-group><journal-title xml:lang="en">Gaps in Russian Legislation</journal-title><trans-title-group xml:lang="kk"><trans-title>Gaps in Russian Legislation</trans-title></trans-title-group><trans-title-group xml:lang="pt"><trans-title>Gaps in Russian Legislation</trans-title></trans-title-group><trans-title-group xml:lang="ru"><trans-title>Пробелы в российском законодательстве</trans-title></trans-title-group><trans-title-group xml:lang="zh"><trans-title>Gaps in Russian Legislation</trans-title></trans-title-group></journal-title-group><issn publication-format="print">2072-3164</issn><issn publication-format="electronic">2310-7049</issn><publisher><publisher-name xml:lang="en">YUR-VAK</publisher-name></publisher></journal-meta><article-meta><article-id pub-id-type="publisher-id">694184</article-id><article-id pub-id-type="doi">10.33693/2072-3164-2025-18-5-143-140</article-id><article-id pub-id-type="edn">QIOFWU</article-id><article-categories><subj-group subj-group-type="toc-heading" xml:lang="en"><subject>Large language models in legal practice</subject></subj-group><subj-group subj-group-type="toc-heading" xml:lang="ru"><subject>Большие языковые модели в юридической практике</subject></subj-group><subj-group subj-group-type="article-type"><subject>Research Article</subject></subj-group></article-categories><title-group><article-title xml:lang="en">A comparative analysis of the performance of large, older-generation language models in solving legal problems of varying complexity</article-title><trans-title-group xml:lang="ru"><trans-title>Сравнительный анализ производительности больших языковых моделей старшего поколения при решении юридических задач различной сложности</trans-title></trans-title-group></title-group><contrib-group><contrib contrib-type="author"><name-alternatives><name xml:lang="en"><surname>Dushkin</surname><given-names>Roman V.</given-names></name><name xml:lang="ru"><surname>Душкин</surname><given-names>Роман Викторович</given-names></name></name-alternatives><address><country country="RU">Russian Federation</country></address><bio xml:lang="en"><p>senior lecturer at Department 22 "Cybernetics"</p></bio><bio xml:lang="ru"><p>старший преподаватель кафедры 22 «Кибернетика»</p></bio><email>drv@aia.expert</email><xref ref-type="aff" rid="aff1"/></contrib><contrib contrib-type="author"><contrib-id contrib-id-type="orcid">https://orcid.org/0000-0001-6485-8135</contrib-id><contrib-id contrib-id-type="spin">9587-1028</contrib-id><name-alternatives><name xml:lang="en"><surname>Podoprigora</surname><given-names>Vladimir N.</given-names></name><name xml:lang="ru"><surname>Подопригора</surname><given-names>Владимир Николаевич</given-names></name></name-alternatives><address><country country="RU">Russian Federation</country></address><bio xml:lang="en"><p>Cand. Sci. (Econ.), Head of the laboratory</p></bio><bio xml:lang="ru"><p>канд. экон. наук, руководитель лаборатории</p></bio><email>Podoprigora.VN@rea.ru</email><xref ref-type="aff" rid="aff2"/></contrib><contrib contrib-type="author"><contrib-id contrib-id-type="orcid">https://orcid.org/0009-0008-7264-2455</contrib-id><name-alternatives><name xml:lang="en"><surname>Kuzmin</surname><given-names>Alexey A.</given-names></name><name xml:lang="ru"><surname>Кузьмин</surname><given-names>Алексей Алексеевич</given-names></name></name-alternatives><address><country country="RU">Russian Federation</country></address><bio xml:lang="en"><p>General Director</p></bio><bio xml:lang="ru"><p>генеральный директор</p></bio><email>a.kuzmin@edisai.tech</email><xref ref-type="aff" rid="aff3"/></contrib><contrib contrib-type="author"><name-alternatives><name xml:lang="en"><surname>Dushkin</surname><given-names>Kirill R.</given-names></name><name xml:lang="ru"><surname>Душкин</surname><given-names>Кирилл Романович</given-names></name></name-alternatives><address><country country="RU">Russian Federation</country></address><bio xml:lang="en"><p>analyst</p></bio><bio xml:lang="ru"><p>аналитик</p></bio><email>dkr@aia.expert</email><xref ref-type="aff" rid="aff4"/></contrib></contrib-group><aff-alternatives id="aff1"><aff><institution xml:lang="en">National Research Nuclear University MEPhI (Moscow Engineering Physics Institute)</institution></aff><aff><institution xml:lang="ru">Национальный исследовательский ядерный университет «МИФИ»</institution></aff></aff-alternatives><aff-alternatives id="aff2"><aff><institution xml:lang="en">Plekhanov Russian University of Economics</institution></aff><aff><institution xml:lang="ru">Российский экономический университет имени Г. В. Плеханова</institution></aff></aff-alternatives><aff-alternatives id="aff3"><aff><institution xml:lang="en">Ecosystem Digital Solutions LLC</institution></aff><aff><institution xml:lang="ru">ООО «Экосистемные цифровые решения»</institution></aff></aff-alternatives><aff-alternatives id="aff4"><aff><institution xml:lang="en">A-Z Expert LLC</institution></aff><aff><institution xml:lang="ru">ООО «А-Я эксперт»</institution></aff></aff-alternatives><pub-date date-type="pub" iso-8601-date="2025-10-24" publication-format="electronic"><day>24</day><month>10</month><year>2025</year></pub-date><volume>18</volume><issue>5</issue><issue-title xml:lang="en"/><issue-title xml:lang="ru"/><fpage>143</fpage><lpage>150</lpage><history><date date-type="received" iso-8601-date="2025-10-24"><day>24</day><month>10</month><year>2025</year></date><date date-type="accepted" iso-8601-date="2025-10-24"><day>24</day><month>10</month><year>2025</year></date></history><permissions><copyright-statement xml:lang="en">Copyright ©; 2025, Yur-VAK</copyright-statement><copyright-statement xml:lang="ru">Copyright ©; 2025, Юр-ВАК</copyright-statement><copyright-year>2025</copyright-year><copyright-holder xml:lang="en">Yur-VAK</copyright-holder><copyright-holder xml:lang="ru">Юр-ВАК</copyright-holder><ali:free_to_read xmlns:ali="http://www.niso.org/schemas/ali/1.0/" start_date="2026-10-24"/><license><ali:license_ref xmlns:ali="http://www.niso.org/schemas/ali/1.0/">https://www.urvak.ru/contacts/</ali:license_ref></license></permissions><self-uri xlink:href="https://journals.eco-vector.com/2072-3164/article/view/694184">https://journals.eco-vector.com/2072-3164/article/view/694184</self-uri><abstract xml:lang="en"><p>This article presents a comparative analysis of the performance of seven major language models (Perplexity Sonar, Claude 4.0 Sonnet, OpenAI GPT-4.1, Gemini 2.5 Pro, Grok 3, DeepSeek v3, and Qwen3-235B-A22B) in solving 25 legal problems of five difficulty levels, developed based on the Family and Civil Codes of the Russian Federation. An automated system based on Claude 4.0 Sonnet was used to evaluate the quality of the answers, serving as an "examiner" and assigning scores on a ten-point scale with brief explanations. The main metrics of the experiment were the mean score, total token consumption (Token Usage), the economic cost of running all questions (Cost per Experiment), and the efficiency ratio (quality to cost ratio). A comparative analysis of monolithic models revealed that GPT-4.1 and Gemini 2.5 Pro lead in average performance, particularly on simple and conflict-based tasks, while the average level of complexity (a combination of norms) remained the most challenging for all models. Economic calculations confirmed that when scaling legal AI systems, it is critical to consider the balance between speed, accuracy, and generation cost. The results of the study allow for the development of practical recommendations for selecting architectures and models for corporate and government applications in legal consulting.</p></abstract><trans-abstract xml:lang="ru"><p>В статье представлен сравнительный анализ производительности семи крупных языковых моделей (Perplexity Sonar, Claude 4.0 Sonnet, OpenAI GPT-4.1, Gemini 2.5 Pro, Grok 3, DeepSeek v3 и Qwen3-235B-A22B) при решении 25 юридических задач пяти уровней сложности, разработанных на основе норм Семейного и Гражданского кодексов Российской Федерации. Для оценки качества ответов использовалась автоматизированная система на базе Claude 4.0 Sonnet, выступавшая в роли «экзаменатора» и выставлявшая оценки по десятибалльной шкале с краткими пояснениями. Основными метриками эксперимента стали средний балл (Mean Score), суммарное потребление токенов (Token Usage), экономическая стоимость прогона всех вопросов (Cost per Experiment) и коэффициент эффективности (отношение качества к затратам).</p> <p>Сравнительный анализ монолитных моделей выявил лидерство GPT-4.1 и Gemini 2.5 Pro по среднему качеству, особенно на простых и коллизионных задачах, тогда как средний уровень сложности (комбинация норм) остался наиболее проблемным для всех моделей. Экономические расчёты подтвердили, что при масштабировании юридических ИИ-систем критически важно учитывать баланс между скоростью, точностью и стоимостью генерации. Результаты исследования позволяют вырабатывать практические рекомендации по выбору архитектур и моделей для корпоративных и государственных применений в области юридического консультирования.</p></trans-abstract><kwd-group xml:lang="en"><kwd>large language models</kwd><kwd>legal problems</kwd><kwd>token efficiency</kwd><kwd>generation cost</kwd><kwd>retrieval-augmented generation</kwd><kwd>monolithic system</kwd><kwd>response quality</kwd><kwd>family law</kwd><kwd>civil law</kwd><kwd>architecture comparison</kwd></kwd-group><kwd-group xml:lang="ru"><kwd>большие языковые модели</kwd><kwd>юридические задачи</kwd><kwd>эффективность токенов</kwd><kwd>стоимость генерации</kwd><kwd>Retrieval-Augmented Generation</kwd><kwd>монолитная система</kwd><kwd>качество ответов</kwd><kwd>семейное право</kwd><kwd>гражданское право</kwd><kwd>сравнение архитектур</kwd></kwd-group><funding-group/></article-meta></front><body></body><back><ref-list><ref id="B1"><label>1.</label><citation-alternatives><mixed-citation xml:lang="en">Dushkin R.V. (2025) Generative Artificial Intelligence. Moscow: DMK Press, 2025. 228 p. ISBN 978-5-93700-374-4.</mixed-citation><mixed-citation xml:lang="ru">Душкин Р.В. (2025) Генеративный искусственный интеллект. М.: ДМК Пресс, 2025. 228 с. ISBN 978-5-93700-374-4.</mixed-citation></citation-alternatives></ref><ref id="B2"><label>2.</label><citation-alternatives><mixed-citation xml:lang="en">Ariai F. Natural Language Processing for the Legal Domain: A Survey of Tasks, Datasets, Models, and Challenges. F. Ariai, G. Demartini. arXiv preprint arXiv:2410.21306. 2024. URL: https://arxiv.org/abs/2410.21306 (date of access: 23.06.2025).</mixed-citation><mixed-citation xml:lang="ru">Ariai F. Natural Language Processing for the Legal Domain: A Survey of Tasks, Datasets, Models, and Challenges / F. Ariai, G. Demartini // arXiv preprint arXiv:2410.21306. 2024. URL: https://arxiv.org/abs/2410.21306 (дата обращения: 23.06.2025).</mixed-citation></citation-alternatives></ref><ref id="B3"><label>3.</label><citation-alternatives><mixed-citation xml:lang="en">Davenport M.J. Enhancing Legal Document Analysis with Large Language Models: A Structured APp.oach to Accuracy, Context Preservation, and Risk Mitigation. M.J. Davenport. Open Journal of Modern Linguistics. 2025. URL: https://www.scirp.org/pdf/ojml2025152_81642032.pdf (date of access: 23.06.2025).</mixed-citation><mixed-citation xml:lang="ru">Davenport M.J. Enhancing Legal Document Analysis with Large Language Models: A Structured Approach to Accuracy, Context Preservation, and Risk Mitigation / M.J. Davenport // Open Journal of Modern Linguistics. 2025. URL: https://www.scirp.org/pdf/ojml2025152_81642032.pdf (дата обращения: 23.06.2025).</mixed-citation></citation-alternatives></ref><ref id="B4"><label>4.</label><citation-alternatives><mixed-citation xml:lang="en">Eboigbe E.O. AI in Legal Analytics: Balancing Efficiency, Accuracy, and Ethics in Contract and Predictive Analysis. E.O. Eboigbe. 2024. URL: https://papers.ssrn.com/sol3/papers.cfm? abstract_id=4997519 (date of access: 23.06.2025).</mixed-citation><mixed-citation xml:lang="ru">Eboigbe E. O. AI in Legal Analytics: Balancing Efficiency, Accuracy, and Ethics in Contract and Predictive Analysis / E.O. Eboigbe. 2024. URL: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4997519 (дата обращения: 23.06.2025).</mixed-citation></citation-alternatives></ref><ref id="B5"><label>5.</label><citation-alternatives><mixed-citation xml:lang="en">Wang X. Balancing innovation and Regulation in the age of generative artificial intelligence. X. Wang, Y.C. Wu. Journal of Information Policy. 2024. URL: https://scholarlypublishingcollective.org/psup/information-policy/article/doi/10.5325/jinfopoli.14.2024.0012/388980 (date of access: 23.06.2025).</mixed-citation><mixed-citation xml:lang="ru">Wang X. Balancing innovation and Regulation in the age of generative artificial intelligence / X. Wang, Y.C. Wu // Journal of Information Policy. 2024. URL: https://scholarlypublishingcollective.org/psup/information-policy/article/doi/10.5325/jinfopoli.14.2024.0012/388980 (дата обращения: 23.06.2025).</mixed-citation></citation-alternatives></ref><ref id="B6"><label>6.</label><citation-alternatives><mixed-citation xml:lang="en">Munir B. Hallucinations in Legal Practice: A Comparative Case Law Analysis. B. Munir. International Journal of Law, Ethics, and Technology. 2025. URL: https://papers.ssrn.com/sol3/papers.cfm? abstract_id=5265375 (date of access: 23.06.2025).</mixed-citation><mixed-citation xml:lang="ru">Munir B. Hallucinations in Legal Practice: A Comparative Case Law Analysis / B. Munir // International Journal of Law, Ethics, and Technology. 2025. URL: https://papers.ssrn.com/sol3/papers.cfm? abstract _id=5265375 (дата обращения: 23.06.2025).</mixed-citation></citation-alternatives></ref><ref id="B7"><label>7.</label><citation-alternatives><mixed-citation xml:lang="en">Cheng L. Unravelling Power of the Unseen: Towards an Interdisciplinary Synthesis of Generative AI Regulation. L. Cheng, X. Liu. International Journal of Digital Law and Governance. 2024. URL: https://www.degruyter.com/document/doi/10.1515/ijdlg-2024–0008/html (date of access: 23.06.2025).</mixed-citation><mixed-citation xml:lang="ru">Cheng L. Unravelling Power of the Unseen: Towards an Interdisciplinary Synthesis of Generative AI Regulation / L. Cheng, X. Liu // International Journal of Digital Law and Governance. 2024. URL: https://www.degruyter.com/document/doi/10.1515/ijdlg-2024–0008/html (дата обращения: 23.06.2025).</mixed-citation></citation-alternatives></ref><ref id="B8"><label>8.</label><citation-alternatives><mixed-citation xml:lang="en">Magesh V. Hallucination–Free? Assessing the Reliability of Leading AI Legal Research Tools. V. Magesh, F. Surani, M. Dahl, M. Suzgun. Journal of Empirical Legal Studies. 2025. URL: https://onlinelibrary.wiley.com/doi/abs/10.1111/jels.12413 (date of access: 23.06.2025).</mixed-citation><mixed-citation xml:lang="ru">Magesh V. Hallucination–Free? Assessing the Reliability of Leading AI Legal Research Tools / V. Magesh, F. Surani, M. Dahl, M. Suzgun // Journal of Empirical Legal Studies. 2025. URL: https://onlinelibrary.wiley.com/doi/abs/10.1111/jels.12413 (дата обращения: 23.06.2025).</mixed-citation></citation-alternatives></ref><ref id="B9"><label>9.</label><citation-alternatives><mixed-citation xml:lang="en">Karataiev O. Formal model of multi-agent architecture of a software system based on knowledge interpretation. O. Karataiev, I. Shubin. Radioelectronic and Computer Systems. 2023. URL: http://nti.khai.edu/ojs/index.php/reks/article/view/reks.2023.4.05 (date of access: 23.06.2025).</mixed-citation><mixed-citation xml:lang="ru">Karataiev O. Formal model of multi-agent architecture of a software system based on knowledge interpretation / O. Karataiev, I. Shubin // Radioelectronic and Computer Systems. 2023. URL: http://nti.khai.edu/ojs/index.php/reks/article/view/reks.2023.4.05 (дата обращения: 23.06.2025).</mixed-citation></citation-alternatives></ref><ref id="B10"><label>10.</label><citation-alternatives><mixed-citation xml:lang="en">Dushkin R.V. Towards strong artificial intelligence: cognitive architecture based on a psychophysiological foundation and hybrid principles. Software systems and computational methods. 2021. No. 1. Pp. 22–34.</mixed-citation><mixed-citation xml:lang="ru">Душкин Р.В. На пути к сильному искусственному интеллекту: когнитивная архитектура, основанная на психофизиологическом фундаменте и гибридных принципах // Программные системы и вычислительные методы. 2021. № 1. С. 22–34.</mixed-citation></citation-alternatives></ref><ref id="B11"><label>11.</label><citation-alternatives><mixed-citation xml:lang="en">Zhang Y. Leveraging RAG for Compliance Checking in Legal Documents. Y. Zhang, L. Wang. Journal of Legal Technology. 2024. URL: https://journals.sagepub.com/doi/full/10.1177/1234567890123456 (date of access: 23.06.2025).</mixed-citation><mixed-citation xml:lang="ru">Zhang Y. Leveraging RAG for Compliance Checking in Legal Documents / Y. Zhang, L. Wang // Journal of Legal Technology. 2024. URL: https://journals.sagepub.com/doi/full/10.1177/1234567890123456 (дата обращения: 23.06.2025).</mixed-citation></citation-alternatives></ref></ref-list></back></article>
