Korean Media: 25% of Top AI Models are Developed in China, "Two-Power Structure" Further Consolidated!

On August 23, the Korean media "Chosun Ilbo" published an article stating that according to information, one-fourth of the world's highest-performing generative artificial intelligence (AI) models are developed by China. As the U.S.-led AI technological hegemony pattern changes, the focus of the AI industry is being redefined, forming a Sino-U.S. two-power structure. Except for France, other countries even find it difficult to enter the top ranks.

According to the "Intelligence Index" recently released by the AI performance benchmarking organization "Artificial Intelligence Analysis," xAI, Elon Musk's company, developed Grok 4, which scored 73 points, taking the first place, followed by OpenAI's o3-pro (71 points), Google's Gemini 2.5 Pro (70 points), and o3 and o4-mini high (both 70 points) tied for third to fifth places. The top five are all developed by American companies.

Sixth place is occupied by DeepSeek's R10528 model (68 points), while Alibaba's Qwen3, MiniMax, Moonshot AI's Kimi K2, and DeepSeek V3 also entered the top twenty. The United States maintains technological leadership with 14 models such as Grok, GPT, Claude, Gemini, and Lama, while China follows closely with 5 models on the list. Mistral AI from France is the only country outside of the United States and China that made the list.

This index is considered the most credible test in the industry, comprehensively evaluating advanced cognitive abilities of AI that cannot be measured by a single benchmark. Among them, MATH-500 (university-level mathematics), Psycode (scientific-based real-time programming assessment), and LiveCodeBench are used as evaluation criteria, used to quantify the "human task execution ability" that goes beyond simple language understanding.

China's entry into the forefront reflects its quantity advantage in AI papers, patents, and talent base. According to the report by Dimensions, as of 2024, China published 23,695 AI-related papers and obtained 35,423 patents, approximately half of the global total, and 13 times the total number of papers from the five countries of the United States, the United Kingdom, Canada, Japan, and South Korea.

Leading Chinese AI startup DeepSeek is achieving results similar to the U.S. model through open-source and low-cost high-performance strategies, thereby expanding its influence. HSBC, Standard Chartered Bank, and Saudi Aramco have already adopted DeepSeek models, and Amazon Web Services is also providing models based on DeepSeek to its customers.

A report released last month by the Belfer Center at Harvard University's Kennedy School analyzed that "China is superior to the United States in AI-related data and human capital." The analysis by the United Nations Industrial Development Organization shows that among the top 100 AI experts globally, 50 are Chinese, with half coming from Chinese enterprises and institutions.

Original: www.toutiao.com/article/1841248298081370/

Statement: This article represents the views of the author.