Naver will release its large language model HyperCLOVA X as open-source software for commercial use starting April 24, 2025. The move marks a major step in expanding Korea’s sovereign AI ecosystem by making powerful generative AI more accessible.
Naver Cloud CEO Kim Yuwon announced that three lightweight versions of HyperCLOVA X—SEED 3B, 1.5B, and 0.5B—will be available for free to both businesses and researchers. These models can now be used for commercial purposes, unlike earlier versions that were restricted to research-only licensing.
This decision aims to lower barriers for small and medium-sized companies that had hesitated due to costs or licensing limitations. It also aligns with Naver’s broader On-Service AI strategy to create practical, easy-to-use AI tools.
Multimodal AI for Text, Image, and Video
Among the newly released models, HyperCLOVA X SEED 3B stands out. It is a multimodal model that can handle text, images, and videos. Naver says the model performs well in tasks like image description, chart analysis, and object recognition.
SEED 3B was tested across nine benchmark tests—including Video-MME, TextVQA-Val, and Korean VisIT-Bench—and scored higher than similar models from US and Chinese tech companies. In some areas, it matched larger models that use more computing resources.
This makes it suitable for practical services like creating ad copy from product pages or building media-rich travel services.
Focus on Lightweight and Reasoning Models
The smaller SEED models—1.5B and 0.5B—are designed for fast processing and lower GPU requirements. These are expected to find wide application in real-world commercial services where cost-efficiency matters.
CEO Kim highlighted Naver’s focus on developing models that can be easily embedded into services. He mentioned that lightweight and reasoning models are in highest demand now.
Naver also plans to release a new reasoning model later this year. It will improve on tasks like math, programming, and automated information search. The reasoning model will be capable of handling complex prompts, integrating APIs, and choosing tools autonomously to complete tasks.
Building Toward AI That Understands Voice
Naver is also working on expanding HyperCLOVA X’s capabilities to include voice. A new voice model will integrate the core knowledge and reasoning functions from the text model into speech. The goal is to support emotionally expressive voice synthesis and more natural conversations between AI and humans.
Future services may feature AI systems that seamlessly switch between text and voice for more fluid interactions.
Strengthening Korea’s AI Ecosystem
Naver emphasized that making these models open-source is part of its broader mission to develop sovereign AI. The company believes strong national capabilities are needed to lead in AI and that collaboration across sectors is essential.
Kim Yuwon stated, “Sovereign AI is a national imperative,” and called for a broader ecosystem that fosters innovation and integrates AI into daily life.
He concluded by saying that these efforts will help Korea build a more resilient and advanced AI infrastructure.
