返回首页
Source: National Business Daily
Based on the latest data from OpenRouter, the total global token volume of large AI model calls reached 36.1 trillion last week (June 1–7), representing a week-on-week increase of 13.5%. This marks seven straight weeks of growth, reflecting sustained rising demand for large AI models worldwide.
Chinese large AI models recorded a weekly call volume of 14.19 trillion tokens, up 27.49% week on week and posting growth for three consecutive weeks. By comparison, U.S. large AI models saw weekly calls drop to 3.2 trillion tokens, a week-on-week decline of 24.53%. Chinese models have outperformed their U.S. counterparts in weekly usage for six weeks running, holding the top spot globally.
The overall growth of global model calls was mainly driven by Chinese offerings, which are gaining deeper traction among global developers.
Notably, four out of the world’s top five large AI models by weekly call volume last week are developed by Chinese companies:
DeepSeek-V4-Flash stayed in first place for the third consecutive week with 3.69 trillion tokens, a 19% week-on-week increase.
Tencent’s Hy3 preview ranked second for three weeks in a row at 2.94 trillion tokens, down 3% week on week.
Newly launched MiniMax M3 claimed third place with 2.5 trillion tokens. According to its developer, it is China’s first large model integrating advanced coding capabilities, a 1-million-token ultra-long context window and native multimodal features.
Xiaomi’s MiMo-V2.5 took fourth place with 2.19 trillion tokens, surging 50% week on week.
Meanwhile, Claude Sonnet 4.6 and Claude Opus 4.7 both dropped out of the top five, with their weekly call volumes falling 9% and 38% respectively. It is the first time in nearly two months that Claude Sonnet 4.6 has slipped outside the top tier.
The Claude series long held a prominent position in the global developer ecosystem. Its decline signals a major shift in the OpenRouter rankings. On one hand, Chinese models are gaining an edge with strong cost performance, high-frequency adaptability and rapid iteration. On the other hand, developers no longer simply opt for the priciest or most powerful models, but strike a balance among performance, cost and task compatibility.
Additionally, Xiaomi’s MiMo-V2.5-Pro, which ranked eighth in the prior week, fell off the list entirely. Its weekly calls once jumped 321% week on week after a substantial price cut, before retreating recently.
In terms of total weekly call volume across product lines on OpenRouter:
MiniMax’s full model lineup accumulated 3.05 trillion tokens, surpassing those of Xiaomi and Tencent.
DeepSeek’s product portfolio hit 6.75 trillion tokens, outperforming Anthropic and Google and retaining the No.1 position for four consecutive weeks.
中国AI大模型调用量连续6周领跑！全球前5中占4席，DeepSeek稳居第1；Claude跌出前5
来源：每日经济新闻
《每日经济新闻》根据OpenRouter最新数据测算，上周（6月1日至7日）全球AI大模型总调用量为36.1万亿Token，较此前一周增长13.5%，连续七周上涨，大模型调用需求仍在持续释放。
其中，上榜的AI大模型中，中国AI大模型周调用量达14.19万亿Token，环比增长27.49%，连续三周实现增长；同期美国AI大模型周调用量为3.2万亿Token，环比下滑24.53%。中国大模型周调用量连续六周超过美国并稳居全球首位。
本周全球大模型调用量继续增长，主要由中国模型放量带动。中国模型正在更深地进入全球开发者调用体系。
《每日经济新闻》注意到，上周，全球调用量排名前五中，前四款均为中国AI大模型。
其中，DeepSeek-V4-Flash连续三周位居榜首，周调用量达3.69万亿Token，环比增长19%；腾讯Hy3 preview连续三周排名第二，周调用量达2.94万亿Token，环比下滑3%；上周刚推出的MiniMaxM3冲进前三，周调用量达2.5万亿Token。据官方介绍，MiniMaxM3是国内首个同时具备“前沿Coding（编程）能力、1M超长上下文、原生多模态”能力的大模型。小米MiMo-V2.5位居第四，周调用量达2.19万亿Token，环比增长50%。
与此同时，Claude Sonnet 4.6和Claude Opus 4.7双双跌出前五，周调用量分别环比下滑9%、38%。其中，Claude Sonnet 4.6近两个月来首次跌出前五。
Claude系列模型此前长期在全球开发者生态中占据重要位置，跌出前五说明OpenRouter榜单结构正在发生变化。一方面，中国模型在高性价比、高频调用和快速迭代上形成冲击；另一方面，开发者在实际调用中并不总是选择最贵或最强的模型，而是在性能、价格和任务适配度之间寻找平衡。
值得注意的是，此前一周排名第八的小米MiMo-V2.5-Pro跌出榜单。该模型在大幅降价后周调用量一度环比增长321%，但本周出现回落。
从旗下产品的周调用量份额来看，《每日经济新闻》注意到，在OpenRouter上，MiniMax旗下模型的周调用总量达3.05万亿Token，超过小米和腾讯。DeepSeek旗下模型的周调用总量达6.75万亿Token，超过Anthropic、谷歌，连续四周位居第一。
Chinese Large AI Models Lead Global Usage for Six Consecutive Weeks: Four in Global Top 5, DeepSeek Remains No.1; Claude Falls Out of Top Rankings
Source: National Business Daily
Based on the latest data from OpenRouter, the total global token volume of large AI model calls reached 36.1 trillion last week (June 1–7), representing a week-on-week increase of 13.5%. This marks seven straight weeks of growth, reflecting sustained rising demand for large AI models worldwide.
Chinese large AI models recorded a weekly call volume of 14.19 trillion tokens, up 27.49% week on week and posting growth for three consecutive weeks. By comparison, U.S. large AI models saw weekly calls drop to 3.2 trillion tokens, a week-on-week decline of 24.53%. Chinese models have outperformed their U.S. counterparts in weekly usage for six weeks running, holding the top spot globally.
The overall growth of global model calls was mainly driven by Chinese offerings, which are gaining deeper traction among global developers.
Notably, four out of the world’s top five large AI models by weekly call volume last week are developed by Chinese companies:
DeepSeek-V4-Flash stayed in first place for the third consecutive week with 3.69 trillion tokens, a 19% week-on-week increase.
Tencent’s Hy3 preview ranked second for three weeks in a row at 2.94 trillion tokens, down 3% week on week.
Newly launched MiniMax M3 claimed third place with 2.5 trillion tokens. According to its developer, it is China’s first large model integrating advanced coding capabilities, a 1-million-token ultra-long context window and native multimodal features.
Xiaomi’s MiMo-V2.5 took fourth place with 2.19 trillion tokens, surging 50% week on week.
Meanwhile, Claude Sonnet 4.6 and Claude Opus 4.7 both dropped out of the top five, with their weekly call volumes falling 9% and 38% respectively. It is the first time in nearly two months that Claude Sonnet 4.6 has slipped outside the top tier.
The Claude series long held a prominent position in the global developer ecosystem. Its decline signals a major shift in the OpenRouter rankings. On one hand, Chinese models are gaining an edge with strong cost performance, high-frequency adaptability and rapid iteration. On the other hand, developers no longer simply opt for the priciest or most powerful models, but strike a balance among performance, cost and task compatibility.
Additionally, Xiaomi’s MiMo-V2.5-Pro, which ranked eighth in the prior week, fell off the list entirely. Its weekly calls once jumped 321% week on week after a substantial price cut, before retreating recently.
In terms of total weekly call volume across product lines on OpenRouter:
MiniMax’s full model lineup accumulated 3.05 trillion tokens, surpassing those of Xiaomi and Tencent.
DeepSeek’s product portfolio hit 6.75 trillion tokens, outperforming Anthropic and Google and retaining the No.1 position for four consecutive weeks.
中国AI大模型调用量连续6周领跑！全球前5中占4席，DeepSeek稳居第1；Claude跌出前5
来源：每日经济新闻
《每日经济新闻》根据OpenRouter最新数据测算，上周（6月1日至7日）全球AI大模型总调用量为36.1万亿Token，较此前一周增长13.5%，连续七周上涨，大模型调用需求仍在持续释放。
其中，上榜的AI大模型中，中国AI大模型周调用量达14.19万亿Token，环比增长27.49%，连续三周实现增长；同期美国AI大模型周调用量为3.2万亿Token，环比下滑24.53%。中国大模型周调用量连续六周超过美国并稳居全球首位。
本周全球大模型调用量继续增长，主要由中国模型放量带动。中国模型正在更深地进入全球开发者调用体系。
《每日经济新闻》注意到，上周，全球调用量排名前五中，前四款均为中国AI大模型。
其中，DeepSeek-V4-Flash连续三周位居榜首，周调用量达3.69万亿Token，环比增长19%；腾讯Hy3 preview连续三周排名第二，周调用量达2.94万亿Token，环比下滑3%；上周刚推出的MiniMaxM3冲进前三，周调用量达2.5万亿Token。据官方介绍，MiniMaxM3是国内首个同时具备“前沿Coding（编程）能力、1M超长上下文、原生多模态”能力的大模型。小米MiMo-V2.5位居第四，周调用量达2.19万亿Token，环比增长50%。
与此同时，Claude Sonnet 4.6和Claude Opus 4.7双双跌出前五，周调用量分别环比下滑9%、38%。其中，Claude Sonnet 4.6近两个月来首次跌出前五。
Claude系列模型此前长期在全球开发者生态中占据重要位置，跌出前五说明OpenRouter榜单结构正在发生变化。一方面，中国模型在高性价比、高频调用和快速迭代上形成冲击；另一方面，开发者在实际调用中并不总是选择最贵或最强的模型，而是在性能、价格和任务适配度之间寻找平衡。
值得注意的是，此前一周排名第八的小米MiMo-V2.5-Pro跌出榜单。该模型在大幅降价后周调用量一度环比增长321%，但本周出现回落。
从旗下产品的周调用量份额来看，《每日经济新闻》注意到，在OpenRouter上，MiniMax旗下模型的周调用总量达3.05万亿Token，超过小米和腾讯。DeepSeek旗下模型的周调用总量达6.75万亿Token，超过Anthropic、谷歌，连续四周位居第一。