Superb Platform and ZERO Vision Foundation Model

Superb AI is a South Korean industrial vision AI company founded in 2018 and based in Gangnam-gu, Seoul, with offices in San Mateo, California and Tokyo. It addresses the long data preparation and training cycles that slow the adoption of computer vision in factories, logistics sites, security operations and mobility programs.

Its stack combines Superb Platform, which links data curation, labeling, model training, deployment and monitoring in one workflow, with ZERO, an industry-tailored vision foundation model that recognizes new defect types or objects from text, box or sketch prompts without pre-training. Superb Video Analytics adds natural-language search and real-time analytics over camera footage, and models can run in the cloud, on-premise or at the edge. Customers listed include Samsung, LG Electronics, Hyundai Motor Company, Toyota, POSCO and Nippon Steel.

• Superb Platform: one workflow for data curation, labeling, training, deployment and monitoring.
• ZERO foundation model: detects new objects or defects from text, box or sketch prompts without retraining.
• Superb Video Analytics: natural-language search and real-time analytics over camera footage.
• Defect detection: vision inspection tuned for manufacturing quality control.
• Edge and on-premise deployment: models run close to cameras on industrial sites.
• Data labeling services: managed annotation backed by more than 130 million industrial data points.

Electronics and component plants in Vietnam can add new defect classes to visual inspection without collecting thousands of labeled images. Steel, energy and shipbuilding sites can monitor abnormal worker behavior and safety zones from existing cameras. Logistics hubs and airports can search large volumes of video in natural language to investigate incidents.

The company states that ZERO reaches high performance with about 1/100 of the usual training data and 1/10 of the training resources, cutting AI adoption timelines on industrial sites from months to days. The site cites 98% defect detection accuracy, a 10x reduction in labeling effort and a 92.7% abnormal-behavior detection rate in deployments. Video Analytics searches about 10 million frames in under 2 seconds on average.

Vision foundation models are shifting industrial computer vision from per-project model training toward prompt-based setup that needs far less labeled data. Manufacturers and logistics operators in Asia, including the Korean and Japanese groups now relocating production to Southeast Asia and Vietnam, are looking for inspection and safety AI that can be rolled out quickly across many lines and sites.

Founded in 2018 and headquartered in Seoul, with entities in the United States and Japan. Total funding is about USD 43 million, with investors including Y Combinator, Samsung NEXT, Hyundai, Doosan, Kakao Investment and POSCO. Named AWS Rising Star Partner of the Year (2024) and selected for Korea's Sovereign AI Foundation Model project. Holds 81 patents and shows ISO 27001 and SOC 2 marks.