Chinese labs release open AI models as US access tightens

In short: Chinese AI companies are releasing strong AI models with open weights, while US models are becoming harder to access and US officials are raising alarms.

What's going on

Several Chinese AI labs have recently released new AI models, including Z.ai’s GLM 5.2 in June, Moonshot AI’s Kimi K3 in mid-July, and Alibaba’s Qwen 3.8 this week. These models are being shared with “open weights,” which means the core model files are available for others to download and run on their own computers (like getting the recipe, not just ordering the dish).

Outside tests suggest some of these models perform close to top US models on popular tasks, especially coding and web development. One ranking site, Arena AI, places Kimi K3 at the top for web development tasks. Artificial Analysis, another benchmarking group, ranks K3 third on its intelligence index. Moonshot AI also temporarily limited new sign-ups after high demand.

At the same time, access to some leading US models has become more restricted. The article describes Anthropic limiting access to its Mythos model and OpenAI delaying GPT 5.6 after a White House request. US officials have also raised concerns about Chinese models. White House science adviser Michael Kratsios alleged Moonshot AI “distilled” Anthropic’s model, meaning it may have learned from another company’s system in a way the US considers improper. Moonshot AI did not immediately respond, according to Wired.

What to watch

More people and startups may try these open Chinese models as alternatives, especially if US models stay locked down. Watch for possible US government actions, including closer scrutiny or sanctions, and for how companies handle claims about copying and intellectual property.

Source: Wired