Llamarine: Open-source Maritime Industry-specific Large Language Model
To address the insufficient domain-specific reasoning capabilities of general-purpose large language models (LLMs) in maritime navigation, this paper introduces Llamarine—the first open-source large language model specifically designed for marine navigation. Methodologically, we employ continual pretraining and supervised fine-tuning on a high-quality, domain-specific corpus comprising maritime textbooks, peer-reviewed academic papers, and authoritative online resources, and propose the first dedicated benchmark for maritime decision-making evaluation. Key contributions include: (1) releasing the first open-source foundational maritime LLM; (2) establishing a systematic, high-fidelity maritime corpus infrastructure; and (3) introducing a reproducible domain-adaptation paradigm. Experimental results demonstrate that Llamarine significantly outperforms both general-purpose and commercial LLMs on trajectory planning, collision-risk assessment, and compliance with international maritime regulations—validating the effectiveness and necessity of domain-specialized modeling.