Korea Deep Learning Wins Global AI Benchmark for Document AI

Korea Deep Learning's proprietary Document AI has achieved the number one overall ranking on ParseBench. This is a global benchmark for document parsing, organized by LlamaIndex. The company secured the top spot in the VLM category with a score of 76.4. They also led in Visual Grounding with a score of 78.8. This win was driven by their KDL-Frontier-Parser-nano, a 1.2-billion-parameter single model. This compact model delivered superior accuracy against larger systems at a fraction of the cost. Korea Deep Learning's technology helps minimize error rates and extracts data with context-aware precision. Their solutions support secure deployments and are trained on over 400 million images and documents. They work with over 80 major corporations and the Gyeonggi Provincial Government, cutting document processing times by up to 96%. This victory follows their KDL Frontier model clinching the number one spot in the English category of OCRBench v2 earlier this year. The company's CEO, Jihyun Kim, was also named to Forbes Asia's 2026 '30 Under 30' list. This shows how specialized AI can outperform broad, general-purpose large language models in specific business tasks.