Naver has released a lightweight artificial intelligence (AI) model that can be used across various defense settings, including drones, unmanned systems, and tactical vehicles. Having established a dedicated organization for defense AI transformation (AX), Naver is now fleshing out its defense AI full-stack strategy by equipping itself with a proprietary defense-specific model.
Naver Cloud announced Thursday that it unveiled its self-developed lightweight omnimodal AI model, 'HyperCLOVA X SEED 4B,' at the '2026 Korea Institute of Military Science and Technology General Conference' held on Jeju Island from the 10th to the 12th of this month.
At the event, Naver Cloud presented various defense AI use cases, including real-time analysis in edge environments such as drones and tactical vehicles. It also unveiled a defense AX roadmap based on sovereign AI that can be operated securely even in closed-network environments, introducing a defense AI full-stack strategy spanning infrastructure, AI models, and AI agents.
HyperCLOVA X SEED 4B, unveiled this time, is a lightweight omnimodal model optimized for defense environments. Equipped with self-developed vision and audio encoders, it can simultaneously process various types of data, including video, audio, and text. In particular, the company explained that by learning Korean documents and domestic environmental data, the model has secured comprehension capabilities suited to the Korean military's operating environment.
Because data security and real-time performance are key to defense AI, the model must run stably not only in the cloud but also in limited computing environments such as drones, unmanned systems, and tactical vehicles. Naver Cloud optimized an existing 8B-class (8 billion parameters) large language model (LLM) through pruning and knowledge distillation techniques, reducing the model size by about half. On the other hand, Naver Cloud emphasized that by improving the omnimodal training process, performance was enhanced beyond that of the existing 8B model. As a result, low-latency inference becomes possible even in edge environments, significantly increasing applicability in defense settings, according to the assessment.
HyperCLOVA X SEED 4B can be used in various fields, including object detection based on drone and coastal surveillance footage, change detection in satellite imagery, identification of hazards in shooting ranges and barracks, automatic recognition of military equipment, and battlefield map analysis. It is also expected to be applied to ISR (intelligence, surveillance, and reconnaissance) automation, explainable unmanned systems, and the establishment of integrated situational awareness systems.
"We plan to build the foundation for defense AX by the end of this year and participate in various defense projects," a Naver Cloud official said. "By 2030, we will expand AI agents across all defense domains and advance autonomous operational support systems to complete the AI transformation in the defense sector step by step."
