Chinese open-weight models now process more than three times the weekly token volume of their US counterparts on the popular AI aggregation platform
For most of 2025, US-developed AI models were the undisputed kings of OpenRouter, the model aggregation platform that lets developers route API calls across dozens of AI systems. They commanded nearly 70% of top-model usage share. That era is over. Chinese AI models have not just caught up. They’ve lapped the competition, processing roughly 18 trillion tokens per week on the platform compared to about 5.5 trillion for US models. That’s a greater than 3-to-1 ratio, and the gap keeps widening.
The crossover point and what followed #
The inflection point arrived during the week of February 9-15, 2026. That week, Chinese models processed 4.12 trillion tokens on OpenRouter, edging past the 2.94 trillion handled by US models. It was the first time Chinese systems outpaced American ones on the platform.
The platform’s total weekly volume has ballooned to over 25 trillion tokens, up from around 5 trillion just six months prior. Most of that explosive growth has been absorbed by Chinese models, not American ones.
Why developers are switching #
Chinese open-weight models from firms like DeepSeek, Qwen, MiniMax, and Moonshot/Kimi have carved out a reputation for competitive pricing and rapid iteration cycles.
Open-weight models are AI systems where the model weights are publicly available, allowing developers to download, modify, and deploy them without paying per-token fees to a centralized provider. This stands in contrast to the closed API model favored by US leaders like OpenAI and Anthropic, where every token processed comes with a price tag.
OpenRouter’s own analysis, drawn from over 100 trillion tokens processed on the platform, confirms this pattern. The demand for AI inference isn’t just growing. It’s shifting geographically, with developers worldwide increasingly routing their workloads through Chinese-built systems.
Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our