What happened
Chinese AI models, including DeepSeek, MiniMax, and Xiaomi MiMo-V2.5, have surpassed US counterparts like ChatGPT and Google Gemini in global token consumption, per OpenRouter data. DeepSeek V4 Flash led with 4.63 trillion tokens, followed by MiniMax M3 at 4.13 trillion, and Xiaomi MiMo-V2.5 at 3.8 trillion. This shift, driven by lower token pricing and energy efficiency, places Chinese models prominently in the top-10 token usage. Google Gemini 2 and OpenAI's GPT 5.5 ranked 12th and 13th respectively.
Why it matters
Cost-efficiency now dictates frontier model adoption, shifting the competitive landscape for platform engineers and procurement teams. Chinese models offer significantly lower token pricing. This economic advantage drives higher usage, while US firms like Uber and Microsoft are capping internal AI usage due to budget overruns. Uber and Microsoft are grappling with AI budget overruns. Procurement teams must re-evaluate model sourcing based on token economics, not solely on raw performance.
