China's new large language model (LLM) is making waves globally with its impressive performance. Concerns are rising domestically that the advancement of China's AI industry may provoke new sanctions from the United States.
The Beijing-based AI startup Moonshot (Chinese name: 月之暗面) unveiled its new LLM model, Kimi 3, on July 17, receiving widespread acclaim from global media.
Kimi 3 Surpasses Anthropic in Coding Abilities
Kimi 3 boasts an open-source model with a staggering 2.8 trillion parameters, achieving performance levels close to the top-tier closed models from the U.S. Despite being open-source, it has garnered high praise for its capabilities.
The model can process documents containing 700,000 to 800,000 words at once. Notably, its coding abilities in the front-end domain have been rated higher than those of Anthropic's latest model. Additionally, Kimi 3 is equipped with AI agent capabilities, enhancing its proficiency in complex programming and research tasks.
Details about the AI chips used for training Kimi 3 have not been disclosed. However, industry speculation suggests it was trained on NVIDIA chips rather than domestic Chinese chips. Nonetheless, Kimi 3 operates within a Chinese data center built with domestic semiconductors.
Moonshot claims to have improved GPU usage efficiency through its proprietary technology. Kimi 3 consists of 896 computing modules, with only 16 activated during actual inference, allowing it to achieve high efficiency even with lower-spec AI semiconductors.
AI Industry Reaches Another Inflection Point
Western media outlets have praised Kimi 3, with Reuters stating that it is evidence of China's AI rapidly closing the gap with the U.S. and that Chinese technology is advancing much faster than expected.
Axios, a U.S. media outlet, noted that the U.S. can no longer confidently claim overwhelming superiority over China in the AI sector, suggesting that China's open-source strategy could pose a threat to American AI companies.
IT publication Tom's Hardware described Kimi 3's coding abilities as among the best in the world, highlighting its impressive support for a context of one million tokens.
Business Insider, an American economic publication, characterized Kimi 3 as a powerful model for businesses, marking it as another inflection point in the AI industry.
China Warns Against Overinterpretation of AI Advances
In China, there is a buoyant atmosphere following Kimi 3's global recognition, reminiscent of last year's Deepfake shock. However, there are also voices cautioning against excessive optimism.
The state-run Global Times published an editorial on July 20, stating, "Moonshot's achievement is certainly something to be proud of," but cautioned against interpreting it as evidence that "China has surpassed the U.S." It added that the U.S. still maintains significant advantages in key areas such as cutting-edge AI models, semiconductor ecosystems, talent, commercialization, and monetization structures.
Furthermore, the Global Times warned that claims of China's AI soon surpassing the U.S. could be used as justification for tightening U.S. semiconductor export restrictions and technology blockades. It emphasized that the development of Chinese AI should not be swayed by exaggerated praise or overestimation from external sources.
As Kimi 3 gains global attention, user demand surged, prompting Moonshot to temporarily halt new user registrations. On July 19, the company announced that "the number of user requests has far exceeded expectations, nearing the capacity limits of the existing cluster," and stated that new registrations would be allowed soon as additional computing resources become available.
* This article has been translated by AI.
Copyright ⓒ Aju Press All rights reserved.