Seong Nak-ho, head of AI technology at Naver Cloud, speaks at the first presentation of the independent AI foundation model project at COEX Auditorium in Seoul on the 30th. /Courtesy of Shim Min-gwan

We will open an era of omnimodel artificial intelligence (AI) that understands the world.

Chief AI technology officer Seong Nak-ho of Naver Cloud stated accordingly on the 30th at the "independent AI foundation model project" first briefing held at COEX Auditorium in Samseong-dong, Seoul, as he unveiled the project blueprint. An omnimodel AI means a single integrated AI model handles multiple data modalities such as text, images and audio. In other words, rather than linking together AI models built for each function, it is designed so that one model can understand, reason and respond across multiple modalities (the forms in which information enters and exits AI).

Seong emphasized that "Naver Cloud focused this project not on a race to scale up models, but on building a 'Sovereign AI' ecosystem that ties infrastructure, talent, data and services into one strand to balance and structure growth."

Naver Cloud first put the inference competitiveness of "HyperCLOVA X SEED" at the forefront. The latest model, "HyperCLOVA X SEED 32B Think," scored 87.4% on "τ2-Bench Telecom," which evaluates composite task performance based on telecom call-center inquiry scenarios. Seong said, "It ranked first within the small-model group of the same size, and it posted a top-tier score in the overall rankings, including major large models," adding, "A model's competitiveness is not the number of parameters but high-quality reasoning ability."

That day, Naver Cloud unveiled "HyperCLOVA X SEED 8B Omni," Korea's first "Any-to-Any" omnimodel in the multimodal domain. Seong said, "Rather than the existing approach of attaching a separate OCR (optical character recognition) model, it can integratively understand diverse modalities such as text, images and audio and enable real-time interaction through 'holistic learning.'" The company said it secured cost efficiency and practicality that can reduce meaning loss in the graphics recognition process while lowering the expense of deployment when applied on industrial sites.

Naver Cloud also presented a plan to raise the user experience by combining speech recognition with inference performance optimized for Korean. As the foundation of this competitiveness, Naver Cloud cited the "Full Stack AI Value Chain." Seong said, "Based on our own AI data center, GPU infrastructure and cloud platform, we vertically integrated the entire process from data collection and generation to model development and service implementation to boost speed and efficiency."

On talent development, Seong said the company is deepening research into core Multimodal AI technologies by running a residency program with research teams at five major universities: KAIST, Pohang University of Science and Technology POSTECH, Seoul National University, Korea University and Hanyang University. He said, "Through cooperation with specialized corporations such as Twelve Labs, we are strengthening our data processing capabilities, while also running a collaborative framework to invigorate Korea's startup ecosystem."

That day, Naver Cloud also said it will accelerate expansion beyond general-purpose models to industry-specific "vertical AI." Seong said, "We will distribute lightweight models optimized for key industries such as finance, defense, manufacturing, retail, defense industry, heavy industry, energy and agriculture to drive AX (AI transformation) and raise national competitiveness."

Naver Cloud's Sovereign AI strategy extends overseas. Seong said, "Based on the implementation experience Naver Cloud has accumulated in Korea, we are building localized AI ecosystems in countries such as Japan, Thailand and Saudi Arabia that reflect each country's cultural and linguistic characteristics." In Japan, Naver Cloud is contributing to solving social issues such as preventing lonely deaths through the "Naver Care Call" service for seniors living alone, and in Thailand and Saudi Arabia it is also pursuing cooperation on local-language large language model (LLM) development and Digital Twin-based Smart City projects.

Seong said, "It is our mission to reach those marginalized by AI first, lead the industry's AX (AI transformation) and safeguard national AI sovereignty," adding, "Centering on the independent foundation model, we will connect infrastructure, talent, industrial application and global expansion into one ecosystem to realize human-centered, inclusive AI in everyday life."

※ This article has been translated by AI. Share your feedback here.