
Alibaba Cloud in China has launched over a hundred open-source models, with its flagship product upgraded to rival GPT-4o in fields such as mathematics

Alibaba Cloud in China released over 100 open-source models on September 19th, and upgraded its flagship product Qwen-Max, claiming that its capabilities in inference, mathematics, and other fields can rival OpenAI's GPT-4o. The newly released Qwen 2.5 model has parameter scales ranging from 500 million to 72 billion, supporting AI applications in areas such as automotive, gaming, and scientific research. In addition, Alibaba Cloud also introduced the Wensheng video model, which can generate videos in multiple visual styles. The company has also comprehensively upgraded its AI infrastructure, improving computational resource management efficiency
BEIJING, Sept 19 (Reuters) - Alibaba Cloud, a subsidiary of China's Alibaba (9988.HK) (BABA.N), released over 100 open-source models on Thursday and upgraded its flagship model product, claiming that its capabilities in inference, mathematics, and other fields can rival OpenAI's GPT-4o. Alibaba Cloud also launched a series of innovative cloud products and services including computing and data center architecture.
In a press release, Alibaba Cloud stated that the newly released Qwen 2.5 open-source model has parameter scales ranging from 500 million to 72 billion, adding over 100 open-source models to support the widespread application of AI in various fields such as automotive, gaming, and scientific research.
Alibaba Cloud also announced an upgrade to its self-developed flagship model, Qwen-Max. The upgraded Qwen2.5-Max performs "equally outstanding" in areas such as language understanding and inference, mathematics, and programming compared to other cutting-edge models (including OpenAI's GPT-4o).
For B-end users, Alibaba Cloud's full-stack AI infrastructure has been upgraded, covering aspects such as green data center architecture, data management, model training, and inference. The company claims that with the latest self-developed cloud-native scheduling engine, model training and inference are integrated to enhance computing resource management efficiency, with the scheduling engine achieving an effective computing power utilization rate of over 90%.
In terms of multimodal products, Alibaba Cloud also introduced a text-to-video model. This model can generate high-quality videos in various visual styles ranging from realistic scenes to 3D animations.
After OpenAI's release of the text-to-video model Sora at the beginning of the year sparked discussions, several Chinese companies have also entered this field. Bytedance's one-stop AI creation platform "DreamAI" developed under its Kuaishou brand was launched in early August.
Bytedance is the latest player to join the wave of generative artificial intelligence in China. In mid-May, the company launched the large model product "Dou Bao" with a focus on low prices. Following Bytedance's aggressive new product pricing strategy, major Chinese large model developers including Alibaba Cloud, Baidu (9888.HK), and Tencent (0700.HK) subsequently significantly reduced the prices of their large model products.
Alibaba executives stated in August that they are confident that the revenue from Alibaba Cloud's customers outside the Alibaba Group will return to double-digit growth in the second half of the fiscal year and accelerate gradually. The Cloud Intelligence Group, which includes Alibaba Cloud, recorded revenue of 26.55 billion yuan in the quarter ending June, a 6% year-on-year increase; revenue from AI-related products continued to achieve triple-digit year-on-year growth. (End)
(Reporting by Lin Yuhang; Editing by Wu Yunling)
