Chinese AI models now claim over 30% of US OpenRouter traffic, and the price gap explains everything
Cost advantages of 60-90% over Western rivals are pulling US developers toward Chinese open-weight models at a pace few saw coming
Something shifted in the American AI developer market around February 8, and the numbers make it hard to ignore. Chinese-built AI models crossed the 30% threshold of weekly token usage among US users on OpenRouter, the popular API aggregation platform that lets developers swap between dozens of large language models.
That number has held above 30% every week since, peaking at 46%. To put the trajectory in context: these same models averaged roughly 11% of US token usage across the prior year.
From footnote to force
The climb started well before February. A joint analysis by OpenRouter and Andreessen Horowitz traced Chinese open-source models from approximately 1.2% of platform token usage in late 2024 to nearly 30% at various points in 2025, based on a sample of 100 trillion tokens.
By the week of February 24, the trend reached a new watermark. Chinese models captured 61% of total token volume among the top 10 models on the platform that week.
Leading that surge was MiniMax M2.5, which logged 2.45 trillion tokens used during the period. Moonshot AI’s Kimi K2.5 and Zhipu AI’s GLM-5 rounded out the top performers.
OpenRouter functions essentially as a model marketplace, letting developers route API calls to whichever model fits their needs.
The cost gap is doing the heavy lifting
The core story here is not geopolitics. It is margin math. Chinese models are running between 60% and 90% cheaper on a per-token basis compared to equivalent offerings from OpenAI and Anthropic.
The models attracting this volume are open-weight, meaning their parameters are publicly released and can be run independently or accessed through third-party platforms like OpenRouter. This matters because it lowers the switching cost for developers. There is no proprietary lock-in, no enterprise negotiation required. You swap a base URL and an API key, and you are running a different model.
What this means for the broader AI market
When 61% of top-model token volume on a major aggregation platform flows to Chinese alternatives in a single week, that is a signal that performance parity is close enough for most production workloads that price becomes the deciding factor.