Alibaba Cloud, the cloud unit of Alibaba Group, has launched a series of new (artificial intelligence) AI tools and infrastructure upgrades at its annual Apsara Conference. Among the highlights is the open-source release of over 100 large language models from the Qwen 2.5 series, as well as a new text-to-video AI model designed to generate high-quality visuals from text prompts.
The Qwen 2.5 series, featuring models with parameter sizes ranging from 0.5 to 72 billion, aims to strengthen AI’s capacity for tasks involving mathematics, coding, and multi-language support. This release targets diverse applications in sectors such as gaming, automotive technology, and scientific research. In total, Alibaba Cloud’s Qwen models have already recorded more than 40 million downloads across platforms like Hugging Face.
“This initiative will empower developers and companies to further leverage AI technologies,” said Jingren Zhou, CTO of Alibaba Cloud Intelligence.
The company also enhanced its flagship Qwen-Max model, improving its performance in areas such as language comprehension and reasoning.
Aside from the Qwen models, Alibaba Cloud introduced a new text-to-video tool under its Tongyi Wanxiang large model family. This AI model is capable of generating videos from both English and Chinese text instructions, with applications ranging from realistic scenes to 3D animations. The technology, powered by a diffusion transformer architecture, transforms static images into dynamic videos.
The cloud provider also rolled out updates to its vision language model, Qwen2-VL. The model now supports longer video comprehension and is integrated with video-based question-answering capabilities. Alibaba Cloud envisions this model being deployed in mobile phones, cars, and robots to automate tasks such as decision-making and specific operations.
Developer support
To further support developers, the company unveiled AI Developer, a Qwen-powered assistant that automates tasks like code programming and bug identification. This tool aims to reduce the burden on developers, allowing them to focus on more complex responsibilities while the AI handles repetitive tasks.
Alibaba Cloud’s new AI infrastructure is supported by its next-generation data center architecture, CUBE DC 5.0. Designed for improved operational efficiency, the CUBE architecture features advanced cooling systems and modular designs, reducing deployment times compared to traditional builds. The company introduced the Open Lake solution to maximize data utility for AI applications, particularly generative AI (GenAI), by integrating data workflows and optimizing resource use.
During the conference, Alibaba Cloud also debuted its 9th Generation Enterprise Elastic Compute Service (ECS), boasting improvements in speed and efficiency, with the new ECS instances offering a 30% increase in search recommendation speed and enhanced query performance compared to previous generations.