Wen Zheng announced a comprehensive upgrade of Alibaba Cloud's Qwen AI platform, centered on bringing agents into production, adding agent services and industry AI solutions on top of model services, and introducing API Express mode, Agent Studio and Qwen AI cockpit solutions. He also reported that Alibaba Cloud's MaaS customers grew sixfold in a year, compute scale is expected to exceed 20GW by 2032, model cold-start time dropped from 1,200 to 70 seconds, and first-token latency fell 38%.
At the Yunqi Conference on September 22, 2026, Alibaba Group CEO Wu Yongming said that multimodal models integrating understanding and generation are an important research direction, and that AI needs perception and interaction capabilities so people can communicate with it as naturally as with other humans.
Zheng Sishou announced at the Apsara Conference that Qwen will accelerate Personal Agent development, using the agentic capabilities of the Qwen 3.8 series models to build a new Personal Agent form and serve 300 million users, reaffirming the goal of giving everyone an AI assistant.
At the 2026 Apsara Conference on September 22, Ping An Group CTO Wang Xiaohang introduced the company's AI practices in finance and healthcare, as well as its collaboration with Alibaba Cloud. Daily token usage grew from about 30 billion to over 300 billion, compute rose from 800 PFLOPS to 1,500 PFLOPS, token output per unit of compute increased by 368%, and unit token cost fell by 76%. AI now handles 25% of in-hospital and 70% of home-based service intake, and around 80% of new code is AI-generated.
At the 2026 Apsara Conference, Alibaba announced that the Honor Magic9 series, launching on September 28, will be among the first official models to feature its AI phone solution Qwen Intelligence, with the Honor Robot Phone also supporting it. The two companies jointly developed a vertical-domain model and solution based on Qwen Intelligence and Honor MagicOS, achieving a comprehensive task accuracy of 91.8%, a GUI operation speed of 3.6 seconds, and an end-to-end service completion rate exceeding 90%.
At the Apsara Conference, Liu Dayiheng said large models have entered the agent era, where models can autonomously plan, call tools, and complete tasks independently, driving exponential global token demand. He noted Qwen series has over 3 billion downloads, 300,000 derivative models, 460+ open-sourced models, and Qwen 3.8 27B became the most-liked model on HuggingFace within two months, while reiterating the commitment to open-source for universal AI accessibility.
At the 2026 Apsara Conference, Alibaba CEO Wu Yongming said customer AI demand is strong, and the company will fully invest in AI infrastructure, aiming for Alibaba Cloud's global data center capacity to exceed 20GW by 2032. He called AI models, AI chips, and AI cloud the three cornerstones of the machine intelligence era, and predicted machine thinking will handle 99.9% of thinking work in the future.