The landscape of the global AI competition is changing: The open-source approach helps Chinese manufacturers rise and shakes the US advantage.

On August 14th, foreign media such as The Wall Street Journal reported that China's rapid development in the field of open - source artificial intelligence (AI) is attracting wide global attention, especially from Washington and Silicon Valley. They are worried that China may surpass the United States in AI models and are currently actively exploring countermeasures.

This year, China has made rapid progress in the AI field. In January, DeepSeek launched its widely - noticed R1 inference model. Subsequently, domestic manufacturers such as Alibaba's Qwen model, Darkside's kimi, Minimax, and Zhipu also stepped up their efforts, releasing a series of new products since July. These models all adopt an open - source approach, allowing users not only to download and use them but also to modify and optimize them, which provides an opportunity for the rapid expansion of China's open - source AI technology.

Foreign media reported that, in contrast, US companies that have long maintained closed - source codes are facing great pressure. At the beginning of August this year, OpenAI released its first open - source model, GPT - oss, which also reflects the US response to the open - source trend.

Technological history shows that many technological fields have gone through a process from a competitive landscape to oligopoly, such as the Windows operating system, Google Search, and mobile operating systems like iOS and Android. However, the winner that ultimately becomes the industry standard may not be the one with the most advanced technology. Ease of access and flexibility are often the key factors in determining victory. This is one of the reasons why the United States is uneasy about China's progress in the open - source AI field.

In the report "Winning the Competition: A US Artificial Intelligence Action Plan" released on July 24th, the Trump administration stated that open - source models "may become the global standard in certain business areas and academic research" and called on the United States to deploy "leading open - source models based on American values" as soon as possible.

Currently, the direct commercial returns of open - source AI are limited because of its high development costs and difficulty in direct monetization. However, companies that successfully attract and retain users can profit from value - added services in the future — just as Google provides revenue - generating services such as Search and YouTube through the Android system (which is itself based on open - source Linux).

Open - source has always been regarded by researchers as an effective way to promote technological innovation. It allows users to view and improve source codes, accelerating technological progress. Foreign media pointed out that in China, the government not only supports the development of open - source AI but also encourages open - source R & D in fields such as operating systems, semiconductor architectures, and engineering software.

Currently, open - source AI is being increasingly adopted by commercial institutions. Many customers prefer open - source solutions because they can freely adjust the models and deploy them on their own systems, ensuring the security and controllability of sensitive information. For example, OCBC Bank in Singapore has developed about 30 internal tools using open - source models, such as using Google Gemma to summarize documents, Qwen to assist in programming, and DeepSeek to analyze market trends.

OCBC Bank said that its strategy is to avoid relying on a single model, continuously monitor the release of new models, and be ready to switch at any time. Executive Director Donald MacDonald said, "We are currently using about 10 open - source models." They prefer models familiar to the developer community for easier technical support.

Data from research firm Artificial Analysis shows that since November 2024, China's top open - source models have outperformed their US counterparts in tasks such as mathematics and programming. Alibaba's Qwen3 has even outperformed OpenAI's GPT - oss in some tasks.

However, foreign media reported that Chinese models are generally larger in size. For example, Qwen is almost twice the size of OpenAI models, which means that when dealing with relatively simple tasks, Qwen may require more computing power to achieve the same effect.

However, engineers, especially those in Asia, reported that Chinese models are generally better at understanding local languages and cultural nuances. Because their training data contains more Chinese content, they also perform better in dealing with some other Asian languages.

Japanese engineer Usami Shinichi from Yokohama recently chose Alibaba's Qwen model when developing a customer - service chatbot for a retail client. He said, "When using top - tier US models, we found that the chatbot sometimes had difficulty capturing the implied meaning of users' words, and the responses were sometimes inappropriate; while Qwen seems to be better at handling these subtleties."

China's AI industry initially focused on price competition among closed - source models. In recent years, the competition has spread to the open - source field, with manufacturers competing for users and public recognition.

Shanghai - based technology analyst Charlie Chai (from 86Research) said, "Chinese companies usually value user stickiness more than short - term profits."

Analysts pointed out that startups currently have a window period to attract users, but this will not last. Ultimately, large technology companies with a large user base are more capable of making profits by providing value - added products such as professional applications or cloud services.

Andrew Ng, the head of Silicon Valley startup DeepLearning.AI, recently commented, "This fierce competition will eliminate many existing players, but it will also forge true winners."

This article is from "Tencent Technology", author: Jin Lu. Republished by 36Kr with permission.