中国民营企业AI大模型集群式崛起，在全球获得广泛认可。根据OpenRouter数据，2026年以来中国AI大模型周调用量稳居全球首位，最近一周达86.74万亿Token，是同期美国的4.5倍，且连续二十四周超过美国1。全球周调用量排名前五均为中国民企AI大模型1。
中国大模型因高性价比和优异性能获得用户高频选择。以月之暗面Kimi K3为例，其每百万Token输入价格为3美元、输出价格为15美元，平均每项任务成本达2美元1。相比之下，Anthropic的Claude Fable5.1输入价格为10美元、输出价格为50美元，平均每项成本达7.63美元1。Claude Fable5.1平均每项成本是DeepSeek V4.1 Flash的28倍1。截至2026年3月，美国顶尖模型和中国顶尖模型之间的差距仅有2.7%1。英特尔CEO陈立武表示中国前沿AI大模型性能已足够好，能满足实际应用需求1。
中国AI大模型正通过开源和商业合作赋能全球开发者和企业。DoorDash、爱彼迎、西门子等企业纷纷采用中国AI大模型1，并通过开源为东南亚、欧美、非洲等地的开发者和企业提供支持1。
Chinese private-sector artificial intelligence large models have achieved widespread international adoption, with their usage surging significantly compared to competitors. According to OpenRouter data, since 2026, Chinese AI large models have maintained the world's leading position in weekly token invocations, reaching 86.74 trillion tokens in the most recent week—4.5 times the volume of the United States—and have consistently exceeded American models for twenty-four consecutive weeks.1
The competitive advantage of these models stems from their combination of cost efficiency and performance quality. Moonshot AI's Kimi K3 offers input pricing of $3 per million tokens and output pricing of $15 per million tokens, with an average task cost of $2, whereas Anthropic's Claude Fable 5.1 is priced at $10 for input and $50 for output, averaging $7.63 per task—making Claude Fable 5.1 approximately 28 times more expensive than DeepSeek V4.1 Flash per task.1 All five of the top-ranked global models by weekly invocations are Chinese private-sector offerings.1 As of March 2026, the performance gap between leading American and Chinese models stands at only 2.7 percent.1
The appeal of Chinese AI models extends across major international enterprises and diverse geographic regions. Intel CEO Chen Li-wu has noted that China's frontier AI large models possess sufficient performance capabilities to meet practical application requirements.1 Major companies including DoorDash, Airbnb, and Siemens have increasingly adopted Chinese AI large models for their operations.1 These models have also been made available through open-source initiatives, enabling developers and enterprises across Southeast Asia, Europe, the Americas, and Africa to benefit from this technology.1
评论
还没有评论，欢迎留下第一条。