A wave of powerful new open-source AI models from Chinese firms, including Moonshot AI's Kimi K3, Z.ai's GLM 5.2 and Alibaba's Qwen 3.8, has drawn immediate attention from Silicon Valley and Washington, echoing the shock caused by DeepSeek's R1 model in January 2025. The releases underscore a widening divergence between Chinese and American approaches to AI development, with Chinese labs embracing open weights while US firms increasingly restrict access to their most advanced systems.

Third-party benchmarks show the new Chinese models performing nearly as well as top Western ones, with a particular focus on agentic coding tasks. The response from US officials has been sharp: adviser David Sacks called Kimi K3's performance "concerning", Commerce Secretary Scott Bessent floated possible sanctions on Chinese AI firms, and White House science adviser Michael Kratsios accused Moonshot AI of distilling Anthropic's Fable model to build K3, calling it "unacceptable". Meanwhile, Anthropic and OpenAI have both faced government-driven delays or restrictions on releasing their own frontier models (Mythos and GPT 5.6 respectively), while Chinese firms such as Alibaba have reaffirmed their commitment to open-sourcing models like Qwen.