On November 26, according to the Financial TimesChinaAlready globally”Open Source”AI ModelsMarket transcendsUSAThis has provided a key advantage in the global application of this powerful technology。
A study by the United States MIT and Open Source AI Entrepreneurship, Hugging Face, found that over the past year, China ' s share of newly developed open source models in total downloads of open source models had risen to 17%, surpassing United States companies like Google, Meta and OpenAI by 15.8%. This is the first time that Chinese companies have surpassed their American counterparts in this area。
OPEN SOURCE MODELS ALLOW DEVELOPERS TO DOWNLOAD, MODIFY AND INTEGRATE FREE OF CHARGE, MAKING IT EASIER FOR START-UPS TO BUILD PRODUCTS AND RESEARCHERS TO IMPROVE TECHNOLOGY. THE WIDESPREAD DISSEMINATION OF SUCH MODELS WILL HAVE A SIGNIFICANT IMPACT ON AI ' S FUTURE DEVELOPMENT。
The Chinese initiative to push the open source model stands in stark contrast to the “closed” model of most technology giants in the United States. Companies such as OpenAI, Google and Anthropic prefer to have complete control over their state-of-the-art technology and to benefit from user subscriptions or business cooperation. In contrast, Chinese companies open up model privileges。
Wendy Chang, a senior analyst at the China Institute in Mercato, says:“In China, the open source model is more mainstream than in the United States. American companies chose not to do so. They rely on high valuations obtained in closed mode to make money and do not want to open up their own business secrets.”
According to data from MIT and Hugging Face, DeepSeek and Ali Baba's mega-models account for the vast majority of downloads from open source models in China。
In January this year, DeepSeek shocked Silicon Valley when he launched his powerful AI reasoning model R1. Despite the use of minimal cost and calculus, the model performed in the same way as the United States competitors. R1 release raises a series of questions: Can the U.S. AI Laboratory, with more resources, hold the competitive edge? Is it worth investing billions of dollars in data centres to run powerful models
In its AI Action Plan, the United States Government has tried to convince United States enterprises to invest in open-source models that are consistent with United States values. In August this year, OpenAI released the first "open source weight" model. Such models can be accessed free of charge, but their availability is maintained compared to the “open source” model, which provides training codes and complete data。
The Llama series model of Meta has been the pole of the open source weight AI model for many years, but the company is shifting its strategic focus to the development of a more robust closed model, competing for “super smart” heights with firms like OpenAI, Anthropic and Google。
China’s Strength
Shayne Longpre, a researcher at MIT involved in the study, noted that Chinese companies such as DeepSeek and Aliyun had pioneered a model release. He indicated that Chinese companies were continuously rolling out models at a weekly or biweekly pace, offering multiple models for users to choose, in contrast to the semi-annual or annual release of a series of models by American AI。
ACCORDING TO OTHER EXPERTS, ALTHOUGH UNITED STATES EXPORT CONTROLS FOR HIGH-END CHIPS LIMIT CHINA ' S ABILITY TO CALCULATE, CHINA HAS A LARGE AMOUNT OF INDIGENOUSLY TRAINED RESEARCH PERSONNEL. THIS ADVANTAGE ENABLES CHINA AI TO USE MORE CREATIVE MODELS TO DEVELOP PATHWAYS, SUCH AS DISTILLATION TECHNOLOGY TO BUILD SMALLER BUT MORE PERFORMANCE MODELS. IN ADDITION, THE AI LABORATORY IN CHINA HAS DEVELOPED THE AI VIDEO GENERATION MODEL。
At the same time, AI in the United States focused more on advancing the state-of-the-art “frontier” model. OpenAI and Google DeepMind are working to develop universal artificial intelligence (AGI) beyond human intelligence。
There are far fewer large, independent developers in the open source field in the United States compared to China. The latest developments in the open source field in the United States came from the Allen Institute of Artificial Intelligence in Seattle, which released a fully open source Olmo 3 model in November。
Janet Egan, Senior Fellow and Deputy Director of the New American Security Center of the think tank, stated: “China has made significant progress in the field of open source models, which should be of interest to the United States.”