Domyn-Large

Domyn Large is a 260B parameter large language model designed for efficient and effective reasoning in constrained environments. It supports extended multi-step reasoning, long-document comprehension, and tool-use workflows. The model can operate with reasDomyn Large is a 260B parameter large language model designed for efficient and effective reasoning in constrained environments. It supports extended multi-step reasoning, long-document comprehension, and tool-use workflows. The model can operate with reasoning on to enhance performance in complex tasks requiring detailed analysis, or off for faster, more compact outputs. Optimized for mathematics and coding, Domyn Large combines strong reasoning capabilities with a practical deployment footprint: a single node with 8x NVIDIA H100 GPUs is sufficient for inference. Domyn Large supports more than 50 languages, with high proficiency in European languages such as Italian, Spanish, French, and German, and strong capabilities in Japanese, Russian, Chinese, Arabic, Hindi, Indonesian, Korean, and more. It delivers strong performance with contexts up to 128K tokens, enabling extended reasoning and comprehension over very long documents. Development included supervised fine-tuning (SFT) to strengthen instruction following and ensure consistent performance across tasks. The model achieves state-of-the-art results on enterprise-critical benchmarks including Text-to-SQL, Text-to-Cypher, knowledge graph extraction, and safety classification, making it well-suited for regulated industries use-cases such as financial services, defense, and advanced manufacturing.