Chinese privately developed AI models are winning users worldwide on price-performance, with fresh data showing the gap in cost between Chinese and American frontier models has become one of the biggest factors reshaping the global AI market.
Figures from aggregate platform OpenRouter showed that in the week of October 5-11, Chinese AI models processed 86.74 trillion tokens, 4.5 times the 18.97 trillion logged by American models, and occupied all five spots at the top of the global usage leaderboard.
On price, US research firm Artificial Analysis calculates that Moonshot AI's Kimi K3 charges 3 dollars per million input tokens and 15 dollars per million output tokens, with an average task cost of 2 dollars, while Anthropic's Claude Fable 5.1 charges 10 and 50 dollars respectively, averaging 7.63 dollars per task. That makes the American model 3.3 times more expensive per token and 3.8 times per task. Claude Fable 5.1's per-task cost is 28 times that of DeepSeek V4.1 Flash.
Performance is closing too. Stanford University's 2026 AI Index Report found that top Chinese and American models have repeatedly swapped the lead on performance benchmarks, with the gap at just 2.7 percent as of March 2026.
Intel's chief executive recently said Chinese frontier models are not only cheap to use but good enough for real applications, with market adoption accelerating. Analysts attribute the cost advantage to algorithmic innovation that cuts compute and inference expense, plus stable energy supply and mature computing infrastructure.
Source: China News Service, October 18-19, 2026.
