#How Did Chinese AI Models Surpass US Models?
Chinese AI models have significantly overtaken US-developed systems in terms of usage on OpenRouter, the model aggregation platform. For the majority of 2025, US AI models held a dominant position, commanding nearly 70% of the platform's top model usage. However, that changed drastically as Chinese models accelerated their growth. Currently, they process approximately 18 trillion tokens each week, while US models account for only about 5.5 trillion, illustrating a ratio exceeding 3-to-1.
#What Was the Turning Point for AI Models?
The pivotal moment occurred in the week of February 9-15, 2026, when Chinese models processed 4.12 trillion tokens, surpassing the American counterparts at 2.94 trillion tokens. This marked the first instance where Chinese models outperformed US models on this platform. Furthermore, the total weekly volume of tokens processed soared to over 25 trillion from just around 5 trillion six months prior, with Chinese models absorbing most of this growth.
#Why Are Developers Choosing Chinese Models?
Developers are increasingly favoring Chinese open-weight models from leading firms like DeepSeek, Qwen, MiniMax, and Moonshot/Kimi. These systems are gaining attention due to their competitive pricing structures and rapid development cycles. Open-weight models allow developers to access publicly available model weights, enabling them to download, modify, and deploy the models without incurring per-token fees associated with centralized providers—unlike the closed API models maintained by US leaders such as OpenAI and Anthropic.
In summary, as indicated by OpenRouter's analysis of over 100 trillion processed tokens, the demand for AI inference is not only increasing but is also shifting globally. Developers are increasingly routing their workloads through systems built in China, emphasizing a significant shift in the competitive landscape of AI technology.