China’s Open AI Models Disrupt Silicon Valley’s Strategy

The artificial intelligence sector may not be witnessing a full-blown DeepSeek 2.0 revolution, but the momentum is unmistakable. China’s top AI laboratories have unleashed a wave of nearly frontier-level open-source models in rapid succession. In June, Z.ai dropped GLM 5.2; last week, Moonshot AI launched Kimi K3; and this Monday, Alibaba released Qwen 3.8.

These developments have sparked immediate discussion in Silicon Valley and Washington, with K3 drawing the most attention as the standout performer. David Sacks, a venture capitalist and AI adviser to President Donald Trump, described Moonshot’s model performance as “concerning.” Commerce Secretary Scott Bessent even hinted at possible U. S. sanctions targeting Chinese AI firms. On Wednesday, White House Office of Science and Technology Policy Director Michael Kratsios leveled a serious accusation, claiming the Trump administration has “information that Moonshot AI distilled Anthropic’s Fable for the development of its K3 model.” He characterized this as “stealing proprietary US technology and undermining American research,” calling it “unacceptable.” Moonshot AI did not immediately respond to a request for comment.

The new Chinese models share several traits. According to third-party benchmarks, they perform nearly on par with the best Western counterparts. They are also optimized for agentic coding tasks, which have become the hottest trend in AI this year. Crucially, they are released,or will soon be released,with open weights, ensuring accessibility and transparency.

Perhaps the most significant parallel to January 2025, when DeepSeek’s R1 model stunned the world, is how this moment underscores the diverging paths of American and Chinese AI labs on the issue of openness versus closure. DeepSeek originally challenged the notion that only closed-source models, built on billions of dollars in compute and training investment, could achieve frontier performance. Yet Western labs have largely stuck to their approach. Today, American frontier models feel more restricted than they did a year ago. Anthropic spent months claiming its latest Mythos model was so dangerously adept at hacking that only approved collaborators could access it. When it finally reached a wider audience, the White House imposed broad export controls, forcing Anthropic to temporarily take Mythos and its less capable sibling, Fable 5, offline. Similarly, OpenAI delayed the release of GPT 5.6 after a White House request.

In China, the landscape looks starkly different. Startups and tech giants have doubled down on open source. Anyone with a sufficiently powerful computer can now download an open-weight model, run it locally, customize it, and enjoy far more freedom than OpenAI or Anthropic would ever permit. In many ways, the open versus closed debate has become more entangled with the US versus China rivalry than ever before.

Chinese labs have compelling reasons for adopting an open-source business strategy. As smaller, newer players in the AI field, offering free and open models helps them attract users, collaborators, and media attention. It also places them in a distinct competitive lane, separate from the deep-pocketed giants like OpenAI, Anthropic, Google, and SpaceX. Earlier this year, rumors surfaced that Alibaba might pivot to closed-source development after reorganizing its AI model teams. But on Monday, the tech giant reaffirmed its commitment by releasing the latest Qwen version,beloved by the global tech community,with open weights, signaling to customers and the public that it is not abandoning the open-source path just yet.