Khaberni - The launch of two new artificial intelligence models by two Chinese companies has sparked a new wave of debate in Washington about how close Beijing is to narrowing the gap with the United States in the AI race, amid American accusations that Chinese companies are leveraging American model technologies to develop their products, according to "Foreign Policy".
The Chinese company "Moonshot AI", based in Beijing, released its new model "Kimmy K3", asserting that it consistently outperforms all competing models except the latest models from "Open AI" and "Anthropic". Two days later, "Alibaba Group" announced similar results for its upcoming model "Queen 3.8".
"Kimmy K3" received significant demand that led the company to suspend new subscriptions within less than 48 hours of its launch due to pressure on its computational capabilities, which analysts considered a proof of the constraints still faced by Chinese AI companies in accessing the necessary infrastructure to run advanced models.
Ryan Fedasiuk, a researcher at the "American Enterprise Institute," described the launch of "Kimmy K3" as a "serious development" and could be considered a "second DeepSec moment", but he also pointed out the structural limitations still facing the Chinese AI sector.
The way Chinese companies have overcome these limitations has alarmed U.S. officials; "Open AI" and "Anthropic" accused Chinese companies of using distillation technology to train their models relying on advanced American models, allowing them to reduce the need for computational capabilities.
Matt Sheehan, a researcher at the "Carnegie Institute," said that distillation "effectively helps compensate for some of the lack of computational capabilities", adding that this does not deny that Chinese companies possess a significant amount of innovation.
Officials in President Donald Trump's administration on Wednesday accused Moonshot of "stealing" American technology to train its new model.
Michael Kratsios, director of the Office of Science and Technology Policy at the White House, said the administration has information that the company used the Claude Fable model from Anthropic along with processing units from "Nvidia" in an illicit way.
The Under Secretary of State for Economic Affairs Jacob Helberg said that "lying, cheating, and stealing is not innovation", while the Treasury Secretary Scott Bisent hinted at the possibility of including Chinese AI companies on lists that restrict access to American technology.
These statements have sparked speculation about the possibility of the U.S. administration imposing broad restrictions on the use of Chinese AI models, although this faces opposition from emerging U.S. companies that rely on low-cost, open-source Chinese models.
Companies like Moonshot and "Alibaba" adopt the open-weight model approach, which allows users to download and run models freely, unlike the closed models employed by American companies such as "Open AI" and "Anthropic".
Sheehan believes this approach makes Chinese models less expensive and easier to use, whether for small American companies or for developing countries looking for less expensive solutions.
In contrast, during his speech at the Global AI Conference in Shanghai, Chinese President Xi Jinping called for promoting openness and cooperation in the development of artificial intelligence, affirming China's readiness to take more open steps, while 29 countries announced joining the new global organization for cooperation in AI initiated by Beijing.
Mohamed Suleiman, director of studies at McLarty Associates, believes that China's reliance on open-source models could enable broader distribution through data centers, cloud service providers, and users, thereby mitigating the impact of the current constraints on computational capabilities.
He added that this strategy resembles approaches previously followed by China in sectors such as solar energy, electric vehicles, and space, aiming to expand the user base, considering that the current limitations may not persist in the future.
