Beijing, China – Chinese e-commerce giant Alibaba announced on Monday its most powerful and largest artificial intelligence (AI) model to date, the Qwen3.8-Max. This launch comes in direct competition with local rival Moonshot AI, which launched its own model last month. This reflects the ongoing race among China’s major technology companies to establish their leadership in advanced AI systems. Furthermore, these companies are expanding their global reach without incurring significant operational costs.
The Qwen3.8-Max model contains approximately 2.4 trillion parameters, the fundamental numerical settings upon which the model relies to learn from data, recognize patterns, generate responses, and execute complex tasks. In contrast, the Kimi K3 model developed by Moonshot contains 2.8 trillion parameters. While this higher number does not necessarily reflect absolute superiority, it has become a key and closely monitored indicator of the computing power and data resources behind intelligent systems.
Global platforms and multimedia are leading
Alibaba unveiled its new model on the Arena.AI crowdsourced modeling platform. The Chinese model ranked first locally among text models, but trailed slightly behind Cloud Fable 5 and three variants of Anthropic’s Opus. However, in the image and visual analysis category on the same platform, Qwen3.8-Max achieved a notable feat, securing second place globally, behind only Cloud Fable 5.
This system is distinguished by its superior ability to simultaneously handle text, images, and videos. It can process up to one million tokens at a time. This enormous capacity allows the model to handle massive amounts of data simultaneously, including lengthy legal files, large software databases, or hundreds of pages of complex text documents.
Operational efficiency and advanced technical design
The architecture of the Qwen3.8-Max model is based on a “Mixture of Experts” design, a technical approach that distributes workloads among specialized parts of the system instead of running the entire model for every request. Actual operation is limited to using only 95 billion operators at a time. This directly contributes to reducing operational costs and response times. The company noted that the model was able to complete a full software engineering project in just 16 days.
The device is scheduled for official release next week via Alibaba Cloud’s Model Studio platform. It’s worth noting that Chinese technology companies are keen to publish parameter counts to promote the adoption of their open-source models among the developer community. In stark contrast, major US companies like OpenAI, Anthropic, and Google keep the parameter counts of their proprietary models confidential.
