Multiple forums at the World Artificial Intelligence Conference (WAIC) opened in Shanghai on July 18th. Across the conference venues, a consensus was taking shape that the next phase of AI would be defined not by competition over model parameters but by its real-world deployment capabilities. AI is moving beyond chat interfaces and into real-world applications that support high-quality development, spanning foundation models, space computing constellations, overseas business expansion, home applications, warehouse robots and digital humans in livestreaming studios.

At the Foundation Model Architecture Innovation and Ecosystem Co-creation Forum 2026, SenseTime officially unveiled SenseNova U1 Pro, its flagship multimodal foundation model. It is the industry's first foundation model designed for AI agents, built on a native unified architecture integrating understanding, generation and action. The model breaks free from the limitations of the compartmentalized design commonly adopted by conventional multimodal models, connecting the entire workflow from visual perception and content generation to task execution. It supports up to native 8K resolution and is designed for professional applications including urban planning, commercial visual design, and film and television production.

The "Xingshu Plan," led by Fudan University, also made a high-profile debut at this year's WAIC. Its initial constellation adopts a minimalist three-satellite cluster comprising a central computing master satellite, the Fuxi meteorological satellite and a real-time remote sensing satellite. The development of the master satellite incorporates five major disruptive innovations in domestically developed hardware, filling several technological gaps in China.

Tencent has continued to accelerate the development of its full-stack AI capabilities this year. Following a major overhaul of its Hunyuan foundation model, Hy3 recorded more than 68 times the total usage of its predecessor, Hy2, within one week of launch, topping the overall model-usage ranking on OpenRouter. Advances in Tencent's foundation-model capabilities have also accelerated the development of its application portfolio, including embodied AI solutions and WorkBuddy.

直击WAIC：AI走出对话框，进入“上天入地”的真实现场

7月18日的上海，世界人工智能大会（WAIC）的多个论坛开讲。会场内外，一个共识正在形成：AI的下半场，不在参数的“军备竞赛”，而在真实场景的落地能力。从基座模型到算力星座，从企业出海到客厅场景，从仓库机器人到直播间数字人，AI正走出对话框，走进高质量发展的真实现场。

在“基座大模型构架创新与生态合作论坛”上，商汤科技正式发布日日新SenseNova-U1 Pro旗舰多模态基座大模型。这是业内首款以“理解·生成·行动”原生统一架构打造的智能体基座，跳出传统多模态拼接式架构的短板，打通视觉认知、内容生成、任务执行全链路。该模型原生支持8K高清画面输出，覆盖城市规划、商业视觉、影视创作等专业场景。

复旦大学牵头打造的“星枢计划”也在本届WAIC同步重磅亮相。首发星座采用1颗中心算力主星+伏羲气象星+即时遥感星极简星簇布局，在主星研制上具有五大颠覆性国产硬件创新，填补多项国内技术空白。

今年以来，腾讯AI全栈能力持续提速。混元大模型完成重构，上线一周的Hy3总调用量较上一代（Hy2）增长逾68倍，登顶OpenRouter全球大模型调用量总榜。基座模型能力的突破，有效驱动了具身智能与WorkBuddy等应用矩阵的快速迭代。

文｜记者 王丹阳 何晶 沈钊
图｜记者 王丹阳
译｜郑奕玲
英文审校｜盛嘉