Naver Cloud is set to launch an inference-specialized model based on its hyper-scale AI, HyperCLOVA X, next month, aiming to stay competitive in the inference model market sparked earlier this year by DeepSeek’s R1. Naver plans to introduce an AI agent capable of reasoning like a human, leading the establishment of a sovereign AI ecosystem.
HyperCLOVA X achieved a score of 90.1 in SimQA (English), a benchmark for measuring the factual accuracy of AI-generated information, comparable to GPT-4o’s web search model, which scored 90.0. It also exceeded 90 points in Math 500 and HumanEval, indicators for math and coding skills.
Unlike models that rely on learned data to reach conclusions, inference models like HyperCLOVA X can independently solve problems through step-by-step reasoning. Major tech companies like China’s DeepSeek, the U.S.’s OpenAI, and Google have all introduced such models, with South Korean players like LG AI Research and startups such as Upstage and Liner also entering the field.
Naver’s reasoning model aims to deliver an AI agent that autonomously solves problems, generating accurate responses in math and programming, while enhancing capabilities like visual and audio information understanding, automatic web searching, API calling, and data analysis. For example, when asked to recommend family-friendly tourist spots and book well-reviewed accommodations in Jeju’s Seogwipo, the model autonomously devises a response plan, utilizing search and reservation APIs to resolve the inquiry.
Naver is also developing multi-modal models that can recognize voice, enhancing AI agent interaction. These models aim to fluidly handle text and voice input, paving the way for more advanced voice-based AI interactions.
Having introduced South Korea’s first hyper-scale AI model in 2021, Naver is accelerating its AI development. It recently unveiled a new flagship model that enhances performance while reducing costs, with expanded visual capabilities. The creation of HyperCLOVA XSeed, an “ultra-light” multi-modal model for commercial use, was made publicly available as open source, supporting the broader adoption of AI technologies.
Industry experts recognize Naver as a unique company in South Korea that has consistently advanced through every development stage of generative AI, with its accumulated expertise becoming increasingly vital as the pace of advancements in inference models and AI agents accelerates.