采用案例显示性价比驱动的趋势：Cursor 共同创办人 Aman Sanger 表示其 Composer 2 建立在 Moonshot 的开源模型 Kimi K2.5 之上；在 Boston University 的活动中，Massachusetts Institute of Technology 讲师 Andrew Mead 提到，中国开源模型在部分指标上接近美国顶级专有模型但成本仅为一小部分。当 Z.ai（前称 Zhipu）将其 coding 方案价格上调 30% 时，仍有人询问是否应趁早购买，反映需求对价格变动仍具韧性；另有 Harvard University 研究者描述「混合」流程：用 Anthropic 的 Claude 做架构，再以更便宜的 Z.ai 的 GLM 反复执行以节省总成本。
量化数据显示使用正在上升但多数仍难以统计：OpenRouter 的资料显示，中国模型使用占比自 2024 年接近 0% 跃升至 2025 年 11 月超过该平台流量的 30%；同时美国领先者仍主导总量，但 Xiaomi 在某周排名中以接近 21% 的市占居首。Mead 预测前沿模型仅需覆盖约 20% 的日常使用，其余 80% 将走向「成本下探」；Stanford University 的 Hao Zhu 估计美国逾 90% 的学术界人士在研究中使用中国模型，且由于可微调权重的开源／开权重模型多出自中国，美国企业的采用往往低调进行以避免地缘政治风险。
Chinese AI firms (from Hangzhou-based Alibaba and Beijing-based Z.ai to Shanghai’s MiniMax) often release models at midnight in China because it aligns with about 9 a.m. in California, targeting developers across the Pacific. Despite Beijing–Washington tensions and data-security worries, the author found Chinese models discussed in seminars and AI courses at top US universities and embraced by entrepreneurs and academics in San Francisco.
Adoption signals a price-performance shift: Cursor co-founder Aman Sanger said Composer 2 was built on Moonshot’s open-source Kimi K2.5, and MIT lecturer Andrew Mead argued Chinese open-source models can approach top US proprietary models on some measures while costing only a fraction. When Z.ai (formerly Zhipu) raised its coding-plan price by 30%, attendees still asked whether to buy before it rose further; a Harvard researcher described a hybrid workflow—use Anthropic’s Claude for architecture, then run cheaper models like Z.ai’s GLM repeatedly to get results at lower total cost.
Hard counts are elusive because open-source models can be downloaded and hosted anywhere, but platform data show rapid growth: OpenRouter reports Chinese models rising from near-zero usage in 2024 to over 30% of its traffic in November 2025. US leaders still dominate volume, yet Xiaomi topped a weekly ranking with nearly 21% share. Mead forecast frontier models may cover about 20% of daily use while the other 80% becomes a race to the bottom on cost and size; Stanford postdoc Hao Zhu estimated over 90% of US academics use Chinese models for research, and US business adoption often stays under the radar due to geopolitical tension.