[Edaily Reporter Kim Hyun-ah ] SECERN AI, which has been pursuing major projects such as airport Smart Pass using facial recognition technology, has also outperformed global AI models in the field of document recognition. The company is expanding its business scope by applying the visual recognition technology it has accumulated through facial recognition to document understanding and business automation.

CUBOX Co., Ltd(340810)The company announced on the 15th that its document AI engine, “SEEU DOC,” ranked first among 25 models in the September 2026 English task evaluation of the global benchmark “OCRBench v2,” scoring an overall 76.2 points. This was 2.8 points higher than the second-place model (73.4 points).

It outperformed models from major global companies, including Alibaba’s Qwen 3.6 (68.3 points), NVIDIA’s Nemotron 3.5 Super VL (67.4 points), Google’s Gemini 3 Pro Preview (63.4 points), and OpenAI’s GPT-5 (55.5 points).

OCRBench v2 is a benchmark that evaluates not only character recognition in documents and images but also information extraction, document structure analysis, computation, and the understanding and inference of visual information. It utilizes 31 scenarios reflecting real-world environments—such as receipts, tables, charts, and handwriting—and measures performance using confidential evaluation data.

Among the eight subcategories, CUDoc ranked first in three: text recognition (77.9 points), document structuring (42.4 points), and reasoning (74.6 points). Notably, its reasoning score was 7.6 points higher than that of the second-place model (67.0 points). It also ranked among the top performers in the remaining five categories, including information extraction (93.8 points) and text detection (80.6 points).

CU-Doc does not merely read text; it extracts data while preserving complex document structures, such as tables and merged cells. It converts the extracted results into JSON, a structured data format, and provides confidence scores for each item, thereby reducing the burden of manual review. It can process PDFs, Hangul (HWP and HWPX) documents, Word, Excel, and PowerPoint files, as well as scanned images.

SECERN AI is integrating CUDoc into the document processing solution of its AI coding agent. This involves utilizing information extracted from documents as knowledge data for Retrieval-Augmented Generation (RAG) or automatically generating electronic forms. The company plans to expand the document automation market targeting public and financial institutions as well as corporations, and also support Proof of Concept (PoC) using actual customer documents.

SECERN AI is a vision AI specialist company that changed its name from its previous name, CU Box. It has been engaged in facial recognition-based security businesses, such as the Incheon International Airport Smart Pass and facial recognition access control systems for government buildings, and ranked first in the VISA-Kiosk and Paperless Travel categories of the U.S. National Institute of Standards and Technology (NIST) Facial Recognition Vendor Test (FRVT) in 2021.

Nam Un-sung, CEO of SECERN AI, stated, “This result is particularly significant because we competed using the same evaluation criteria as the latest AI models from global Big Tech companies,” adding, “We will expand the vision AI technology we have built in facial recognition to document and language understanding, thereby reducing on-site document-related tasks and driving business innovation for our customers.”