China's Open AI Models Challenge Silicon Valley's Closed Approach
Original: China’s Open AI Models Are Challenging Silicon Valley’s Playbook
Why This Matters
The open-vs-closed AI divide is increasingly mapped onto the US-China tech competition, shaping global access to frontier AI.
Chinese AI labs including Z.ai, Moonshot AI, and Alibaba have released a series of open-weight models in mid-2026 that benchmark near Western frontier performance. U.S. officials raised concerns, with the White House alleging Moonshot AI distilled Anthropic's Fable model to build its K3 model.
A wave of Chinese open-source AI releases is intensifying the open-vs-closed debate in the industry. Z.ai released GLM 5.2 in June, Moonshot AI released Kimi K3, and Alibaba released Qwen 3.8, all of which third-party benchmarks show performing nearly on par with leading Western models. The models are optimized for agentic coding tasks and released with open weights, allowing broad access and local deployment.
The releases sparked immediate reaction in Washington. David Sacks, AI adviser to President Trump, called Kimi K3's performance 'concerning.' Commerce Secretary Scott Bessent suggested possible sanctions on Chinese AI firms. White House OSTP Director Michael Kratsios alleged that Moonshot AI used Anthropic's Fable model to train K3, calling it 'stealing proprietary US technology.' Moonshot AI did not immediately respond to comment requests.
The moment echoes DeepSeek's R1 release in January 2025, which challenged assumptions that only closed, heavily-funded models could achieve frontier results. Since then, the divergence has widened: Anthropic restricted its Mythos model to approved collaborators due to hacking capabilities, then temporarily took both Mythos and Fable 5 offline following White House export controls. OpenAI delayed GPT 5.6 after a White House request. Chinese labs, by contrast, have doubled down on open-source availability, offering models anyone can download, run locally, and customize.