Chinese AI labs keep releasing open-weight models while US firms lock theirs down

 

Chinese AI labs keep releasing open-weight models while US firms lock theirs down

A wave of open-weight AI models from Chinese labs is putting fresh pressure on Silicon Valley and drawing pointed reactions from Washington. In the span of just a few weeks, Z.ai released GLM 5.2, Moonshot AI released Kimi K3, and Alibaba released the latest version of its Qwen line, all with open weights, all benchmarking close to the best Western models, and all optimized for the agentic coding tasks that have become the year's dominant AI trend. K3 in particular has drawn the sharpest attention: Trump AI adviser David Sacks called its performance "concerning," Commerce Secretary Scott Bessent floated the idea of sanctions on Chinese AI companies, and White House science and technology policy director Michael Kratsios alleged that Moonshot AI distilled Anthropic's Fable model to build K3, calling it "stealing proprietary US technology."

The episode is drawing comparisons to the DeepSeek moment of January 2025, and for good reason. It reinforces a widening divergence between how American and Chinese labs think about openness. Anthropic kept its most capable Mythos model restricted to approved collaborators for months over hacking concerns, then briefly had to take both Mythos and Fable 5 offline after White House export controls were issued following a wider release. OpenAI delayed the launch of GPT 5.6 after receiving a White House request. Chinese labs, by contrast, have moved in the opposite direction, betting that free, downloadable, customizable models attract more users, collaborators, and goodwill than the locked-down approach American frontier labs have increasingly adopted.

Alibaba's Monday announcement is the clearest signal yet that this strategy is holding. Earlier this year, rumors had circulated that the company might pivot toward closed-source development after reorganizing its AI teams, but its decision to release Qwen 3.8 with open weights suggests it is not changing course. The open-versus-closed debate, once largely a philosophical argument within the AI community, is now inseparable from the geopolitical competition between the United States and China, and the gap between the two camps appears to be growing.

Read More →