GLM-5.2 Dominates: Chinese AI Models Capture 40% of Tokens

Chinese AI models now handle 40% of all tokens processed on OpenRouter, the world's largest model-routing platform. This is a significant jump from under 2% in late 2024. What's interesting is that GLM-5.2, a Chinese open-weight AI model, launched its companion coding agent ZCode, and developers are already creating "ZCode alternatives" guides. This shows a strong market presence. The shift is visible on OpenRouter, which routes live developer API traffic for over 8 million users. By May 2026, Chinese AI companies had 20 models in the platform's daily top-50 rankings, up from five at the start of 2025. The economics of agentic coding drove this acceleration. Programming workloads grew to over 50% of total OpenRouter token volume by mid-2026. Chinese models offered a structural advantage in unit cost. In February 2026, Chinese AI models surpassed American models in weekly API call volume on OpenRouter for the first time. They processed an estimated 5.16 trillion tokens compared to 2.7 trillion from US providers. The bottom line is that the open-weights license on models like GLM-5.2 does not protect against China's National Intelligence Law for many teams, especially those unable to self-host the model's large data. This means developers need to understand the implications of using these widely adopted models.