Chinese company Moonshot AI released a new version of its Kimi model this week, leading to a perhaps inevitable wave of talk about China and open source AI.
Moonshot said that although the Kimi K3 “still trails the most powerful proprietary models, the Claude Fable 5 and GPT 5.6 Sol”, the new open source model “showed borderline performance across the entire evaluation suite, consistently outperforming the other tested models”. Independent analyzes by Arena.ai and Vals AI he also suggested that the Kimi is competitive with the top frontier models.
The announcement, which coincided with speech by Chinese President Xi Jinping at the AI World Congress in Shanghai, seems to have spooked Wall Street, with Nasdaq down about 1% on Friday as investors sold shares in chip companies such as Nvidia.
Many of the resulting posts from tech insiders will sound familiar to those who remember the conversation after another Chinese company, DeepSeek, released its open-source R1 model in January 2025. Except now, everything seems more acute after the Trump administration’s tariff war with China, repeated battles over the national security threat Anthropic supposedly poses, and finally preparing as big companies.
For example, David Sachs—the Trump administration’s former AI czar and now co-chair of the President’s Council of Advisors on Science and Technology— contrasts Kymi’s progress with the United States “tying knots: politicians and bureaucrats are banning new data centers, piling up state regulations and pushing for new federal agencies to pre-approve border models. That’s how you lose the AI race.” (The news also gave him an excuse to do a digging in Anthropic, calling Claude an example of “wake-up lobotomized models” who are “the enemy of American competitiveness.”)
And former CEO of Uber Travis Kalanick echoed the complaints that the Chinese are “distilling” (ie, training on the results of) American AI models.
“If distillation is not enforced, then everyone should be able to distill from everyone else..otherwise one hand [would be] strapped behind the backs of American models,” Kalanick wrote. (Of course, the American models are also built over the Chinese ones, specifically the Kimi.)
Meanwhile, OpenAI’s chief futures strategy Dean Ball he said that Kimi is “a very good model” whose performance probably cannot be “explained by distillation or anything like that”, adding that he is “personally surprised that the Chinese state continues to allow the open supply of models this good, given the potential risks”.
In fact, Ball suggested that “a possible outcome of a world with an open-weight dominant model is full AI communism,” where AI is viewed as “a ‘public good’ that will ultimately be provided by the state as a kind of ‘digital public infrastructure.’
“That future seems like a dystopian landscape to me, but I’ve never met an open-weight model advocate who doesn’t admit that this is where things end,” Ball said. He even suggested to the Trump administration (for which he worked) will eventually realize that it needs to “create large amounts of regulatory risk around the use of Chinese open weight models”.
“You don’t need to ‘ban open source’ (one of the dumbest patterns in the AI policy debate),” Ball said. “You just have to direct every agency to pass soft law that creates FUD [fear, uncertainty, and doubt]. “Federal Reserve advisory finds backdoors may exist in Chinese AI models”. It doesn’t have to be so well justified. You just create enough regulatory risk that every regulated business backs away.”
However, Shakeel Hashim, editor of the AI-focused publication Transformer, argued that much of the concern is overblownboth because Kimi “probably does not have dangerous cyber capabilities,” and because the Chinese government will face “extremely similar incentives” to restrict open Chinese models once they develop those capabilities.
When you purchase through links in our articles, we may earn a small commission. This does not affect our editorial independence.
