Chinese open-weight AI models are spreading quickly worldwide

In short: Chinese companies are openly sharing the core files of advanced AI models, and those models are being used widely around the world.

What's going on

Several China-based AI labs have released “open-weight” language models, including DeepSeek, Alibaba’s Qwen family, and Moonshot AI’s Kimi. “Weights” are the model’s internal settings, like the finished recipe for how it responds, which lets others run it on their own computers instead of relying on one company’s website.

This open approach is helping Chinese models spread fast. The report cited Chinese open models reaching close to 30 percent of global usage for this type of AI, while large US-made closed models still hold a bigger overall share. In late April 2026, Chinese open models processed trillions of “tokens” (tokens are small chunks of text, like words and pieces of words), approaching the volume handled by top US models.

Analysts say developers are drawn to these models because they can be cheaper to run and easier to adapt for specific needs. The article also points to China’s broader advantages, like access to relatively cheap electricity, large-scale manufacturing, and government support that can make it easier for startups to build practical AI products. Some Chinese companies are also linking AI to real-world services, such as Baidu’s lower-cost robotaxis.

What to watch

More governments and companies may pay attention to open-weight AI because it changes who can build and control AI systems. It could also raise new questions about safety rules and oversight when powerful models can be copied and run almost anywhere.

Source: NYTimes