Naver Cloud aims to boost Korea’s AI ecosystem with open-source models

Naver Cloud CEO Kim Yu-won speaks during a press conference at Naver Square in Gangnam District, Seoul, Wednesday. Courtesy of Naver Cloud

Naver Cloud has introduced its first open-source artificial intelligence (AI) model, HyperCLOVA X SEED, for commercial use.

Emphasizing the importance of sovereign AI, company CEO Kim Yu-won said Wednesday that the new model is part of Naver's efforts to contribute to the expansion of Korea’s AI ecosystem. Sovereign AI refers to a country's ability to control, create and deploy its own AI models within its infrastructure, data, workforce and networks.

“In my opinion, sovereign AI is not just a marketing term but inevitable in our time. As global conflicts intensify, its importance is increasing,” the CEO said during a press conference at Naver Square in Gangnam District, Seoul.

“Releasing the open-source AI model will strengthen Korea's sovereign AI ecosystem. It will be beneficial if companies and the country can prepare themselves and turn it into business opportunities.”

The open-source AI is offered in three lightweight models in different sizes: HyperCLOVA X SEED 3B, HyperCLOVA X SEED 1.5B and HyperCLOVA X SEED 0.5B.

A poster for Naver Cloud's open-source AI model, HyperCLOVA X SEED / Courtesy of Naver Cloud

SEED 3B is a vision language model that can process not only text but also image and video information, with the ability to understand charts, recognize objects and describe photos.

Sung Na-ko, executive director of the Hyperscale AI Model, explained the model is more specialized for understanding the Korean language and Korean culture-related images than other global services.

"It displays an almost overwhelming performance in understanding the Korean language,” he said.

“When comparing average scores across nine benchmarks for evaluating visual information capabilities in Korean language, Korean culture and English, SEED 3B scored higher than similar-sized big tech models from the United States and China.”

SEED 1.5B is a model that understands and generates text. Capable of basic translation tasks between major languages like English and Japanese, it can be used as an auxiliary tool in areas like education and business communication.

The ultra-small SEED 0.5B model can implement conversational interfaces with users in Korean, making it suitable for mobile applications, smart home devices and wearable devices.

Sung noted that the company chose to release lightweight models to meet the needs of businesses seeking to implement AI.

“Generally, AI models tend to perform better as they get larger. However, when trying to apply them to actual services, the first issue encountered is cost. Even if we want to transform all services to ChatGPT-level models, realistically, the operational costs are too high to bear,” he said.

“That's why many companies are focusing on small but high-performing lightweight models. These lightweight models are much more efficient and practical even when applying AI partially. So we plan to continuously improve and update lightweight models."

Kim added that this will serve as a catalyst for small and medium-sized businesses that were hesitant to join Korea’s AI ecosystem.

Naver Cloud is also planning to roll out a reasoning model version of its flagship, HyperCLOVA X, in the coming weeks.

The new model will go beyond simple language generation, including generating precise answers to mathematics and programming-related questions, understanding visual and audio information and automatic web searching.

The company says it will continue to develop HyperCLOVA X, focusing on multimodality and low-cost, lightweight and advanced reasoning.