Bayan-15B: Specialized Arabic Legal Reasoning LLM

Bayan-15B is a 14.8 billion parameter language model developed by Bayan AI, LLC, based on the Qwen2.5-14B architecture. It has undergone Continued Pre-Training (CPT) on a specialized corpus of approximately 190 million tokens, comprising over 900 classical Arabic texts focused on legal theory, interpretive methodology, and jurisprudential reasoning. This domain-adapted model demonstrates deep comprehension of traditional scholarly Arabic writing styles and complex argumentative structures.

Key Capabilities

• Legal Text Analysis: Proficient in understanding and generating classical Arabic legal discourse.
• Interpretive Reasoning: Skilled in analyzing methodological frameworks and interpretive principles.
• Classical Arabic: Offers deep comprehension of traditional scholarly Arabic writing styles.
• Argumentation: Capable of following complex chains of reasoning and evidence-based arguments.

Good For

• Academic research in Arabic legal traditions.
• Analysis of classical interpretive methodologies.
• Arabic NLP applications requiring domain expertise in legal and interpretive fields.
• Educational tools for Arabic legal studies.
• Compliance and advisory systems for Islamic finance.

Limitations

Bayan-15B is highly specialized in classical Arabic legal discourse and is not intended as a substitute for qualified legal or religious experts. It functions best as a research and analysis tool, and users may require domain expertise to accurately evaluate its outputs. The model is released under a CC BY-NC-ND 4.0 license, permitting academic and research use, with commercial use requiring separate licensing.