China's open-source AI models rival Silicon Valley, sparking US sanctions talk

A wave of open-weight Chinese AI models is challenging Silicon Valley’s dominance, echoing the DeepSeek R1 moment from January 2025. In recent weeks, Z.ai released GLM 5.2, Moonshot AI launched Kimi K3, and Alibaba debuted Qwen 3.8—all nearly matching top Western models on benchmarks. K3, the strongest of the bunch, ranks fourth globally in agentic tasks (just behind Anthropic’s Fable and Opus 4.8 and OpenAI’s GPT-5.6) and first in web development on Arena AI. The models are optimized for agentic coding—the hottest AI trend—and released with open weights, enabling local customization and transparency.

The US response has been swift and political. Trump adviser David Sacks called K3’s performance “concerning.” Commerce Secretary Scott Bessent suggested sanctions, and White House OSTP director Michael Kratsios alleged Moonshot distilled Anthropic’s Fable to build K3, calling it “unacceptable.” Meanwhile, the US has tightened export controls on its own frontier models: Anthropic temporarily took Mythos and Fable 5 offline after White House requests, and OpenAI delayed GPT-5.6. Chinese labs remain committed to open source, attracting global developers and media spotlight. Alibaba, despite rumors of pivoting to closed source, reaffirmed open weights for Qwen 3.8. The result: the open vs. closed debate is now deeply entangled with US vs. China competition, with Chinese models proving that open source can achieve near-frontier performance.