The top 3 models on openrouter this week ( Chinese models are dominating!)
For the first time, a single model processed over 3 trillion tokens in one week on OpenRouter.
The AI model usage tracker OpenRouter shows a historic shift, with Chinese-developed models now dominating the top three spots by weekly token volume. Alibaba's Qwen 2.5-72B leads with an unprecedented 3 trillion tokens processed in one week, the first model to cross that threshold. DeepSeek's V3 and V2 models follow, each also exceeding 1 trillion tokens. This marks the first week where multiple models have surpassed the 1-trillion-token mark, a milestone previously held only by xAI's Grok-1.5 Fast last month. The data suggests a rapid adoption of high-performing, open-source Chinese models by developers and companies for cost-effective, large-scale AI inference.
- Alibaba's Qwen 2.5-72B set a new record by processing over 3 trillion tokens in a single week on OpenRouter.
- Multiple models, including DeepSeek-V3 and DeepSeek-V2, also exceeded 1 trillion tokens, a milestone previously rare.
- This represents a major competitive shift, with Chinese open-source models surpassing US counterparts like GPT-4 and Claude in real-world API usage volume.
Why It Matters
This signals a shift in the global AI power balance, with open-source Chinese models becoming the go-to for cost-sensitive, high-volume applications.