DeepSeek is now the most-used model author on OpenRouter. 6.6T tokens last week, ahead of Anthropic at 6.1T and every other lab.
DeepSeek: 6.6 trillion
Anthropic: 6.1 trillion
Google: 3.9 trillion
OpenAI: 2.3 trillion
That means DeepSeek alone is processing nearly 3x as many tokens as OpenAI on the platform.
And it is not an isolated result, Chinese model makers collectively account for roughly 20 trillion weekly tokens in this chart.
Open models are no longer used as a fallback