From DeepSeek to Alibaba's Qwen, from Moonshot's Kimi K3 to MiniMax's M3, from Zhipu's GLM to StepFun's Step 5 Preview, and on to ByteDance's Doubao and Baidu's ERNIE — Chinese private-enterprise AI large models have risen in a cluster and are gaining fans worldwide, China News Service reported on October 18.
According to OpenRouter, a global large-model aggregation platform, since 2026 Chinese AI models' weekly token calls have ranked among the global leaders, with multiple head models long topping the weekly token-use charts. Calculations from OpenRouter data show that in the week of October 5–11, total global large-model calls reached 172 trillion tokens, of which Chinese models accounted for 86.74 trillion — 4.5 times the US figure for the same period. China's weekly call volume has exceeded the US for 24 consecutive weeks, firmly ranking first in the world, with the global top five by weekly calls all being Chinese private AI models.
The core reason for global users' high-frequency choice is high cost-performance. Artificial Analysis estimates Kimi K3's price at 3 dollars per million input tokens and 15 dollars per million output tokens, about 2 dollars per task, while Anthropic's Claude Fable5.1 costs 10 and 50 dollars respectively, about 7.63 dollars per task — dozens of times DeepSeek V4.1 Flash and 28 times its per-task cost. High performance is also key: Intel CEO Lip-Bu Tan recently said China's frontier models are low-cost yet good enough for real applications.
Stanford's 2026 AI Index shows that since early 2025, Chinese and US models have repeatedly swapped the top performance spot, with the gap staying in single digits — by March 2026 only 2.7%. Chinese private models are also going open-source, which Nvidia's Jensen Huang called a catalyst for global progress.
Source: China News Service / China Qiaowang
