A model released Friday by Chinese startup Moonshot AI has fueled buzz about the country’s technological capabilities, as experts said it could rival some of the most advanced offerings from American labs.
Large language models support chatbots and other artificial intelligence tools with their ability to collect large amounts of digital data.
Moonshot AI’s “Kimi K3” is one of several from China that is growing in global popularity thanks to lower costs and source code that programmers can customize.
Shortly after its launch, Kimi K3 topped a leaderboard for AI coding powered by a platform called Arena created by UC Berkeley researchers.
This drew excitement from industry insiders, with some evoking a 2025 release from China’s DeepSeek that shook assumptions about US dominance in AI.
“The K3 chemistry looks really good, closer to the limit yet,” said Ethan Mollick, a University of Pennsylvania professor and a leading voice on AI, at X.
But he “can’t write a good murder mystery (though neither can any other model). That remains the sharpest frontier of AI development, he said.
“Feeling a violent market reaction to KimiK3…similar to DeepSeek’s moment,” wrote tech writer and investor Kevin Xu.
Beijing-based Moonshot AI said the Kimi K3 was the world’s first open-source model of its size.
The more internal variables or parameters a model has, the better it can handle complex requirements, and K3 Chemistry has about 2.8 trillion.
Major US players Anthropic and OpenAI do not provide details on how many parameters their main models have.
– “Border Level” –
“Kimi K3 demonstrated borderline performance in our evaluation suite, consistently outperforming other models tested,” said Moonshot AI.
However, overall performance “still trails the most powerful proprietary models” from Anthropic and OpenAI, the company said.
Hussein Abbass, a professor of computer science at UNSW Canberra, said Kimi K3 appears to be good at coding, “but it is not yet known how competitive it is across the range of tasks expected of establishment models”.
Should America’s rivals be worried? “I wouldn’t say they should worry. But they shouldn’t be calm,” to maintain their advantage, Abbass told AFP.
He added that AI performance is not just about the model, but also the hardware that runs it, along with data centers and supply chains.
Arena ranked Kimi K3 ninth worldwide for text queries. When AFP tested Kimi K3, it generated ideas for technology reporting that compared favorably with responses from other chatbots.
The release of the Kimi K3 follows that of other much-hyped Chinese AI models such as Zhipu AI’s GLM-5.2.
It came during a major technology conference in Shanghai, where President Xi Jinping on Friday called for international cooperation on AI governance.
The United States, which restricts exports to its rival of powerful microchips that can train and run AI, said earlier this year that it was about eight months ahead of China in the strategic field.
But the Trump administration recently delayed the public release of high-level models from Anthropic and OpenAI over concerns they could help hackers break into online systems.
“Post-Kimi K3 and open weights models again approaching the limit, I wonder if Anthropic and OpenAI will be allowed to increase their release cadence by the government,” Mollick wrote.
