Alibaba Group Holding demonstrated the Chinese tech giant’s strong AI skills across models and data-center architecture on Thursday by releasing an open-source AI reasoning model that it claimed outperformed DeepSeek‘s R1.
Following the introduction of its QwQ-32B model, Alibaba’s shares listed in Hong Kong saw a nearly 8.4% increase, closing Thursday at HK$140.80. The South China Morning Post is owned by Alibaba.
According to a blog post by the team behind Alibaba’s Qwen family of AI models, the new model, despite having only 32 billion parameters, matched or outperformed DeepSeek’s R1, which has 671 billion parameters, in domains like mathematics, coding, and general problem-solving.
According to the scientists, a lower number of parameters allows the model to function with fewer computational resources, which promotes broader adoption.
Alibaba chairman Joe Tsai’s recent editorial for the Post, in which he emphasized that real-world applications were crucial to maximizing intelligence in AI model development, is consistent with the lean design of Alibaba’s model.
With Alibaba declaring last month that it will invest US$52 billion on cloud computing and AI infrastructure over the next three years—the largest computer project ever financed by a single private company in China—it also comes at a time when AI use is booming throughout the nation.
Additionally, Alibaba claimed that QwQ-32B fared better than OpenAI’s o1-mini, which was constructed using 100 billion parameters. Hugging Face, the biggest open-source AI model community in the world, offers QwQ-32B.
Similar to how DeepSeek developed its R1 model, the Qwen team credited reinforcement learning approaches for the performance gains of its new reasoning model.
According to the team, these developments “not only pave the way for further innovations in the pursuit of artificial general intelligence, but also demonstrate the transformative potential of reinforcement learning.”
During a recent earnings call, Alibaba CEO Eddie Wu Yongming stated that the company’s main goal was to build artificial general intelligence, which he defined as the moment at which AI could achieve 80% of human skills. This launch comes after his words.