A Chinese AI gains ground on the most powerful models of Claude and ChatGPT

BarcelonaUS large artificial intelligence models, which have comfortably dominated the market since the launch of ChatGPT, no longer seem so unattainable. In mid-June, the Chinese start-up Z.ai, a spin-off from Tsinghua University that has been at the forefront of Beijing's technological strategy, released its latest product, GLM-5.2. The analysis published by the company itself places it at a similar level to the main products of OpenAI and Anthropic laboratories. Market analysts have confirmed its potential; cybersecurity company Semgrep, for example, has stated: "We have another Mythos at home." Experts consulted by ARA acknowledge that an independent general assessment is needed, but indicate that, based on the published specifications, GLM-5.2 would be "no more than 1% below" US competition.

Technology analysts praise the Chinese ecosystem's ability to develop these tools. "They have made a virtue of necessity," comments Aleix Valls, founder of the consultancy WeArtificial, in conversation with this newspaper. The technical impediments that the Chinese innovation ecosystem still faces, with less raw computing power and the most advanced chips blocked by the White House, have in fact served to further improve their technology. According to Josep Curto, founder of Athena Core, Washington's restrictions have spurred Asian ingenuity, and Chinese laboratories have focused on programming "more efficient models," which must run on less powerful infrastructures.

This advance demonstrates, in the expert's view, the excesses of Anthropic and OpenAI. Because they have had access to practically unlimited capital, and have been able to build data centers at will, their models "are pure brute force." "They use the entire network even when they don't need it," he insists. In contrast, Z.ai "changes algorithm pieces that are unnecessary or redundant," and thus needs many fewer computers running at the same time. Thanks to this reduction, GLM-5.2 is substantially cheaper than its competitors, with a price between six and seven times lower than the American ones, according to market sources.

Trump's backfire

According to experts, the evolution of GLM-5.2 is the result of the clash of technological strategies between China and the US. According to Curto, Xi Jinping's government "uses any mechanism" to eliminate any technological dependence on its geopolitical rival. Among other supports, Z.ai has received sustained funding from tech giants like Tencent and Alibaba, which have participated in its various funding rounds since 2020.

Conversely, Valls recalls, the Trump administration interpreted that if it managed to prevent the infrastructure for advanced AI and LLMs from reaching the Chinese market, "they would not be able to achieve" technological sovereignty. A thesis that has proven completely wrong: without Nvidia's most advanced chips, China has connected its artificial intelligence labs with tech companies like Huawei to optimize its circuits for local models.

In fact, the founder of WeArtificial identifies a clear political intentionality in the market access pace of major Chinese products. GLM-5.2, it should be remembered, was available only a few hours after the Trump administration's order to block Mythos and Fable, Anthropic's cutting-edge models, for all users without US citizenship. The same has happened with OpenAI's GPT 5.6, which has been limited to a "small group of trusted companions".

Beijing's roadmap is based on "looking for the weaknesses" of the competition and exploiting them. "Given the arbitrariness that Trump shows, the Chinese present themselves as robust," he observes. Furthermore, the Asian giant's AI "start-ups" do not compete among themselves, unlike those in the United States. In the US, "the pie is not big enough for everyone" in terms of capital, while in China "there is a political directive" that accelerates the entire sector, reflects Curto.