HyperCLOVA XUnclaimed

HyperCLOVA X is NAVER's proprietary large language model, offered for businesses that want to build AI features into their products and services. The page describes it as a model with strong capabilities for complex business problems, and says it can be connected to the wider AI ecosystem to create new user experiences.

The page makes several claims about the model. It says HyperCLOVA X was trained on far more Korean data than GPT-4 and performs well on Korean culture and social norms, on Korean writing tasks such as essays, poems, columns and letters, and on machine translation between Korean and English, Japanese and Chinese. It also says the model is proficient in Korean, English, Japanese and Chinese. Its tokenizer is optimised for Korean, which the page says makes it run up to twice as fast at lower cost than English-centric LLMs. The latest version handles images and audio, and can analyse visual data such as images and tables.

The model family includes several options. HyperCLOVA X THINK is aimed at multimodal reasoning, including STEM problem solving and image-based instructions. The SEED range includes reasoning models at 32B and 14B parameters, an 8B Omni model that handles text, images and audio, a 3B model for visual content, and smaller 1.5B and 0.5B models for conversation and translation on resource-constrained devices such as mobile apps and smart home devices. The page also mentions a DASH series for lightweight, fast responses and says SEED models are open source.

The page lists many customer examples, including a personal news service, a Korean honorifics learning app, a laundry guide built from care labels, a quiz generator, a legal Q&A service, a public data chatbot run by the Seoul Metropolitan Government, and writing and survey tools built by other companies.

Pricing is not stated on the page. Businesses are invited to contact sales.