Alibaba Cloud expands AI reach with Qwen3, Wan2.5, and new infrastructure
Alibaba Cloud, the backbone of Alibaba Group’s digital technology and intelligence, announced its full-stack AI innovations at the Apsara Conference 2025. The updates span from large language models to application platforms and upgraded infrastructure, reaffirming its global leadership in AI development.
The company unveiled Qwen3-Max, its most powerful large language model with over 1 trillion parameters. Available in both Instruct and Thinking modes, the model delivers high performance across coding and agentic benchmarks. Alongside, Qwen3-VL, a vision-language model, and Qwen3-Omni, a multimodal omni-model, extend capabilities to text, image, audio, and video with real-time interaction.
Since 2023, Alibaba has open-sourced more than 300 AI models, recording over 600 million downloads and 170,000 derivative models. Its Model Studio platform has enabled more than 1 million developers and corporates to build with Qwen.
Alibaba also previewed Wan2.5, its new visual-generation model for multimedia content creation. Enhancements in speech recognition, image editing, and video generation strengthen its ecosystem for both enterprise and consumer use.
To simplify enterprise agent deployment, the company introduced Model Studio-ADK and ADP, offering both high-code and low-code frameworks. Enterprises can now create complex AI agents with autonomous decision-making and cross-environment deployment. More than 800,000 agents have already been built through Model Studio.
On the infrastructure front, Alibaba Cloud introduced Vector Bucket for efficient vector data storage, HPN8.0 high-performance networking, AI-powered Cloud Threat Detection, scalable Container Compute Services, and upgrades to PolarDB database. These innovations aim to reduce costs, increase scalability, and enhance security in agentic AI development.
Alibaba’s RMB 380 billion investment over three years will drive these advancements, reinforcing its vision of AI models as the “operating systems of the AI era.”
