China accelerates on artificial intelligence with two new models

Leading the way was Beijing-based Moonshot AI, which announced the Kimi K3 on Friday. According to the company’s internal tests, the model would place ahead of almost all US systems, surpassed only by OpenAI’s GPT-5.6 Sol and Anthropic’s Claude Fable 5, although it was ahead on some specific benchmarks.

Alibaba’s response arrived over the weekend, anticipating Qwen3.8, defining it as one of the most powerful models available today, second only to Fable 5. Both companies are focusing on an element that clearly distinguishes them from the large American laboratories: the choice to make the most advanced models public, rather than keeping them closed. An approach already followed by Meta in the United States, but which in China is becoming a distinctive feature of the entire sector.

Moonshot presents Kimi K3 as the world’s largest open source system, with 2.8 trillion parameters. Alibaba instead declares 2,400 billion parameters for Qwen3.8, calling it a “continuously evolving” model. The number of parameters is a rough indicator of the complexity and scale of a model, but does not in itself guarantee superior performance; Neither OpenAI nor Anthropic make the exact data of their respective flagship systems public.

At the moment it is difficult to accurately assess the real capabilities of the two Chinese models, in the absence of a full release and independent tests. Moonshot has announced that it will release Kimi K3’s full weights, the numerical parameters learned during training, on July 27. Alibaba simply said that Qwen3.8 weights will be available “soon.”

Despite uncertainty about actual results, the announcements have already ignited competition between the United States and China in what has long been called the defining technological challenge of the current era. A buzz that recalls that sparked last year by DeepSeek, when the Chinese company presented a low-cost model capable of competing with the most advanced US systems.

The availability of two advanced Chinese models, downloadable and adaptable by anyone, contrasts with the more cautious approach of US laboratories, which keep their most powerful systems proprietary. An opening that comes as Washington intensifies measures to limit the global spread of the technology: the American government has already used export controls to limit China’s access to the most advanced chips, and has forced Anthropic to withdraw its most capable system from the market, due to fears related to a possible competitive advantage for foreign rivals.