虽然千问和元宝的用户规模仍在增长，但这两款产品在2026年的行业舆论中正逐渐淡出视野。[1]千问月活用户达1.67亿，同比增长5792.9%，从去年第五跃升至第二；元宝重回亿级用户规模。[1]然而相比之下，豆包6月月活达3.82亿，同比增长172.1%，已是千问月活的两倍有余。[1]
这种"静悄悄"的现象背后，反映了AI竞赛格局的深刻转变。[1]行业竞争从上半场的"参数竞赛"进入下半场的"商业落地竞争"，话题定义权正从创业公司向大厂转移。[1]创业公司方面，月之暗面Kimi K3推出参数规模达2.8万亿的开源模型，成为全球参数最大；DeepSeek V4 Flash版本采用峰谷计费模式，非高峰期每百万输出Token仅0.28美元。[1]与此同时，大厂正通过生态协同和场景闭环来重新定义竞争规则。[1]
阿里对千问的布局在持续深化。[1]3月，阿里统一品牌为"千问"，但随后经历了团队调整。[1]在商业应用端，千问在中国一汽大模型商业化和荣耀Robot Phone等项目中有所落地。[1]元宝则在7月上线免费Agent功能，并与京东完成小程序生态打通。[1]腾讯方面在2025年12月升级了大模型研发架构，任命姚顺雨为混元首席AI科学家。[1]
China's artificial intelligence competition is entering a new phase, with Alibaba's Qianwen and ByteDance's Yuanbao maintaining user growth momentum despite reduced public attention. Qianwen's monthly active users reached 167 million, representing a year-over-year surge of 5,792.9 percent and elevating it to the second-ranked position in the market, while Yuanbao recovered to hundred-million-user scale [1]. The shift in focus reflects a fundamental transformation: the industry has moved beyond the "parameter race" that dominated earlier competition into a phase centered on commercial deployment and ecosystem integration [1].
The competitive dynamics now favor tech giants leveraging their integrated business structures. ByteDance's Douban recorded 382 million monthly active users in June, marking a 172.1 percent year-over-year increase—more than double Qianwen's user base [1]. Meanwhile, Alibaba has pursued targeted B2B applications, with Qianwen supporting commercial implementations including China First Automobile Works' large-model deployment and Honor's Robot Phone [1]. Yuanbao separately launched free Agent functionality in July and established ecosystem integration with JD.com's mini-program platform [1]. In December 2025, Tencent restructured its large-model research architecture, appointing Yao Shunyu as Chief AI Scientist for its Hunyuan model, signaling intensified competition among major technology corporations reshaping competitive rules through closed-loop scenarios and synergistic ecosystems rather than parameter scale alone [1].
