The latest large language model (LLM) built in China is impressing scientists with its size and capabilities. Last week, Beijing-based Moonshot AI introduced ‘Kimi K3’, a powerful reasoning LLM that can handle large expanses of text. The company’s own tests found that the K3 can match or outperform rival American models in tasks such as coding and spreadsheet manipulation.

Not since DeepSeek has a Chinese model of artificial intelligence has it caused such a stir. “It’s a game-changer,” says Joel Pearson, a cognitive neuroscientist at UNSW Sydney, Australia, who studies how AI is affecting people’s lives. “People call it the ‘Sputnik moment,’” he adds.

Three days after the model’s release, Moonshot AI said it had paused new registrations on K3 because demand had pushed its system near the limits of its processing capacity.

The model’s launch on July 16 came just before the 2026 World Artificial Intelligence Conference in Shanghai. The conference began with Chinese President Xi Jinping announcing the formal start of a global alliance to create regulation to ensure AI is safe and benefits people. “In China’s view, all countries should take a people-centered approach and develop AI for positive and good,” Xi said at the conference, adding that AI should be an engine for “shared prosperity and common security.”

The timing of Kimi K3’s launch is significant, says Mehwish Nasim, an artificial intelligence researcher at the University of Western Australia in Perth. The model was released just weeks after Claude Fable 5 by US company Anthropic, and days after OpenAI released its latest GPT-5.6 model. “China is signaling its ambition not only to build cutting-edge AI systems, but also to help shape the international AI ecosystem and its governance,” he says.

The largest open weight model

K3 is open weight, like its predecessor models Kimi K2 and DeepSeek, which were developed by a Hangzhou-based company of the same name, meaning that its core components are publicly available and researchers can download and modify them for free, while information about the model’s training is private. It can also be accessed via an ‘application programming interface’, which allows the software to communicate, at a lower cost than proprietary LLMs such as those from OpenAI, Google and Anthropic. Those LLMs are closed weight, meaning they cannot be downloaded or modified.

Moonshot says the weights or parameters will be published on July 27. Rahul Shome, a robotics and artificial intelligence researcher at the Australian National University in Canberra, says consumers, developers and researchers have pushed for AI models to be open weight. K3 is particularly interesting because it would be the largest open weight model, Shome says, with 2.8 trillion parameters and a working memory of one million tokens (the text units used by AI models). However, the model is too large to run on personal devices, so it will likely require significant investment from institutions to run, he adds.

K3’s large working memory means it could remember the contents of thousands of lines of code or an entire book and reduce the likelihood of producing fabricated information called hallucination, says Niusha Shafiabady, a computational intelligence researcher at the Australian Catholic University in Sydney. Shafiabady says she will test K3 herself asking him to summarize the research results.

Closing the gap

Historically, the performance and capabilities of open weight models have lagged behind those of American models. But the gap appears to be the narrowest it has ever been, says Aaron Snoswell, an AI accountability researcher at the Generative AI Laboratory at the Queensland University of Technology in Brisbane.

Pearson claims that Chinese openweight models are changing the way investors and users perceive the value of American frontier models. Google, Anthropic and OpenAI spend billions of dollars training their models compared to the much lower budgets of Chinese companies. Additionally, users could lose access to U.S. models if the government decides to impose restrictions, Pearson says. Last month, Anthropic temporarily disabled its most advanced models, the Fable 5 and Mythos 5, for all users after the US government ordered it to suspend access to foreign citizens. Access to Fable 5 was restored this month, but Mythos 5 is only available to specific US organizations.

UNSW computer scientist Toby Walsh says Chinese AI companies have been able to build cutting-edge AI models despite the United States restricting China’s access to advanced AI chips, citing national security concerns. “One suspects that necessity was the mother of invention here,” he adds.