China News Service, Beijing, October 18 (Reporter Liu Liang) From DeepSeek to Alibaba Tongyi Qianwen; from Dark Side of the Moon's Kimi K3 to MiniMax's M3; from Zhipu's GLM to Step5 Preview of Step Stars, to ByteDance's Doubao and Baidu Wenxinyiyan... For some time, China's private enterprise artificial intelligence (AI) large models have emerged in clusters and have gained fans around the world.
According to statistics from the global AI large model aggregation platform OpenRouter, since 2026, the weekly calls of China's large AI models have been among the top in the world, and multiple leading AI large models have been "dominating" the Token weekly call list for a long time.
According to OpenRouter data calculations, the total number of global AI large model calls last week (October 5 to October 11) was 172 trillion Tokens. Among the large models on the list, the weekly call volume of China's large AI models reached 86.74 trillion Tokens, which was 4.5 times that of the United States during the same period. The number of weekly calls for large AI models in China has surpassed that of the United States for 24 consecutive weeks, ranking first in the world. During the same period, the top five global weekly calls were all large Chinese private enterprise AI models.
The large AI models of Chinese private enterprises are frequently chosen by users around the world. The core lies in their high cost performance.
The official website of Artificial Analysis, an American research institution, estimates that China’s Dark Side of the Moon’s Kimi K3 currently has an input price of US$3 and an output price of US$15 per million Tokens, with an average cost per task of US$2; while the US Anthropic’s Claude Fable5.1 has an input price of US$10, an output price of US$50, and an average cost per task of US$7.63. The input price and output price of Claude Fable5.1 are dozens of times that of DeepSeek V4.1 Flash, and the average cost per item is 28 times that of DeepSeek V4.1 Flash.
High performance is also key to its popularity. Intel CEO Chen Liwu recently said that China's cutting-edge AI large models not only have low cost of use, but also have good enough performance to meet the needs of practical applications, and the market adoption rate is also very fast.
According to Stanford University's "2026 AI Index Report", since the beginning of 2025, Chinese and American models have alternately topped the performance list many times. Over the past year, this gap has fluctuated, but it has always remained in single digits. As of March 2026, the gap between the top model in the United States and the top model in China is only 2.7%.
Industry insiders analyze that China's large AI models can take into account performance and cost, and the core lies in the coordinated efforts of the two aspects. On the one hand, thanks to the innovation of technological routes, computing power consumption and reasoning costs have been greatly reduced; on the other hand, stable energy supply and complete computing power infrastructure have further reduced unit computing power costs.
China’s private enterprise AI large models are “open source” to the outside world. Nvidia CEO Jensen Huang previously pointed out that open source AI large models are a catalyst for global progress, giving all countries and industries the opportunity to participate in this revolution.
In Southeast Asia, China's large AI models use localized language adaptation to solve the pain point of the Western general language model, which is mainly English, being "acclimated" to the local environment. In Europe and the United States, in order to reduce technology costs, companies such as American food delivery giant DoorDash, travel home rental platform Airbnb, and German technology giant Siemens have embraced large Chinese AI models.
In African countries, more and more developers are turning the Chinese open source model into a "toolbox" for local innovation: some use it to optimize business service processes, some develop educational tools for high school students, and some build chatbots for local users.
Li Wei, deputy director of the Institute of Cloud Computing and Digitalization of the China Academy of Information and Communications Technology, pointed out that around the world, people in different industries are re-developing based on China's large models, thus feeding back and inspiring a number of new technologies and applications, greatly increasing the speed of integrating AI into practical applications in various industries.
When top AI capabilities are no longer locked within the walls of a few giants, developers around the world can call or even deploy these models independently at low cost. What China’s large open source AI model brings to the world is not just a technology, but a profound change that promotes equal sharing of the dividends of a new round of technological revolution among all countries.
