China Opens AI Models to Catch Up With "Closed" U.S.; South Korea Faces "Selective Openness" Challenge (Comprehensive)

Why Has China Chosen Open AI?
Divergent Strategies: U.S. Closes While China Opens

Chinese artificial intelligence (AI) companies are taking a contrasting approach to the United States by opening up their frontier (cutting-edge) AI models, diverging from the closed-off services operated by American tech firms. South Korea now faces the challenge of determining the scope of openness for its own emerging frontier AI models, positioned between the opposing strategies of the U.S. and China.

According to the industry on July 21, Chinese tech giant Alibaba has released a preview version of its AI model ‘Qwen 3.8 Max’ and announced its transition to an open weights approach. This model, equipped with 2.4 trillion parameters, discloses the weights formed during the AI training process, enabling companies to run it on their own servers or modify it to suit their specific needs.

Recently, Chinese startup Moonshot AI also unveiled ‘Kimi K3’, an open-source model designed for long-duration work such as processing large-scale software repositories and coding. On July 17 (local time), Chinese President Xi Jinping emphasized during his keynote address at the World Artificial Intelligence Conference (WAIC) that "the advancement of AI should be a symphony of international cooperation, not the solo act of a single nation."

This reflects a strategy aimed at spreading derivative models and services based on Chinese open-source models, thereby seizing leadership in the AI ecosystem and setting global technology standards. Open-source models allow users to directly download, utilize, modify, and validate the model to fit their own data and services. Given their performance comparable to cutting-edge U.S. models, but at a much lower price, they may appeal to cost-sensitive small and medium-sized enterprises (SMEs) and developing countries. Notably, these models offer an alternative to the closed models operated by American companies. From the user's perspective, closed AI models come with risks such as dependence on the platform provider’s policy changes—like price hikes or service suspensions.

Currently, the U.S. frontier AI market is dominated by closed models such as OpenAI’s GPT, Anthropic’s Claude, and Google’s Gemini. While companies like Meta (Llama) and Google (Gemma) previously advanced open-source ecosystems, they have recently maintained closed strategies, recognizing advanced AI technology as a national strategic asset and restricting its export. By securing users based on superior performance, these firms generate profits through tiered pricing according to subscription conditions and usage. By not disclosing weights and key training information, the developers can manage model performance, security, safety, and minimize the risk of technology leaks.

Experts indicate that, should South Korea pursue frontier AI model development, a two-track strategy is necessary, with the scope of openness determined according to technological maturity and intended use.

Park Jin-ho, professor of Computer Science and AI at Dongguk University, noted, "AI models have reached saturation in terms of speed and performance, with fierce competition expected between the U.S. and China. Open-source approaches have the limitation that responsibility can be unclear in the event of accidents or bugs." Lee Jaesung, professor in the AI Department at Chung-Ang University, added, "The U.S. is ahead in AI technology and the market, so there is little incentive to open up core models. In contrast, China has adopted an open-source strategy to expand its market share. For Korea, a closed approach is advisable for sovereign AI such as defense and national security, but general models can be open. By doing so, Korea can establish an alternative ecosystem between the U.S. and China."

This content was produced with the assistance of AI translation services.