On the 16th, NC AI announced the release of four research-oriented open-source models, including the multimodal AI model ‘VARCO-VISION 2.0’, demonstrating its capabilities in Korean-based AI technology.
The VARCO-VISION 2.0 14B model is characterized by its enhanced ability to simultaneously analyze complex documents, tables, and charts, as well as its proficiency in bilingual processing of Korean and English. The 1.7B model is designed with a lightweight structure suitable for use on personal devices such as smartphones and PCs.
The VARCO-VISION-1.7B-OCR, specialized in Optical Character Recognition (OCR), applies an AnyRes resolution splitting input structure, achieving high recognition accuracy even in mixed Korean-English environments and processing high-resolution information. The VARCO-VISION embedding model improves search accuracy by calculating semantic similarity across text, image, and video, showing outstanding performance in the MultiVENT2.0 benchmark as well.
These newly released models are expected to have practical applications across various industries such as finance, education, culture, and manufacturing, in areas like document automation, content creation, and video search, according to the company. NC AI has focused on maximizing data efficiency and designing lightweight models to enhance cost-effectiveness and accessibility to technology.
Lee Yeon-soo, CEO of NC AI, stated, “In the trend of multimodal AI transition, VARCO-VISION 2.0 will be an opportunity to enhance the autonomy of domestic AI technology by combining Korean specialization with global-level technology.”