Alibaba unveils Qwen3 Max Preview, its most powerful AI model yet
Alibaba has unveiled its most powerful AI model yet. Here's why Qwen3 Max Preview is turning heads with its massive scale and capabilities.
The race to build the world's most capable AI models is accelerating, and Alibaba has just raised the stakes.
The Chinese technology giant has introduced Qwen3.8-Max, its largest and most advanced artificial intelligence model to date. With 2.4 trillion parameters, multimodal capabilities and support for a massive 1 million-token context window, the new model highlights Alibaba's ambition to compete with the world's leading AI developers.
A giant leap in model scale
Qwen3.8-Max is built with 2.4 trillion parameters, making it one of the largest AI models announced publicly. Parameters are the values an AI model learns during training. They allow the system to recognise patterns, understand language, generate responses and solve problems.
While a larger number of parameters does not automatically translate into better performance, it often reflects the scale of the training process and the computational resources invested in building the model.
Alibaba's latest release comes close to the size of Moonshot AI's Kimi K3, which contains 2.8 trillion parameters. The comparison illustrates how Chinese AI firms are rapidly closing the gap with some of the world's biggest AI developers.
China's open-weight push — and where Qwen3.8-Max fits
One of the defining characteristics of China's AI industry has been its growing focus on open-weight models.
Unlike closed AI systems, open-weight models make their trained parameters available to developers, enabling them to build applications, customise deployments and conduct further research without starting from scratch. This approach has helped Chinese AI companies gain wider adoption among enterprises, software developers and research institutions.
Alibaba's Qwen family has already established itself as one of the country's leading AI platforms, with many of its smaller models released under open-weight licences. Qwen3.8-Max itself, however, is currently a proprietary preview: Alibaba has said open weights for the 2.4-trillion-parameter Qwen3.8 family are coming, but it has not yet published the weights, a licence or a release date. For now, its contribution to the open-weight ecosystem is a stated intention rather than a download.
Built for text, images and long-form reasoning
Qwen3.8-Max is a multimodal AI model capable of understanding text, images and video within a single system. One of its standout features is support for up to 1 million tokens in a single prompt. Tokens are small pieces of text, such as words or parts of words, that AI models process while generating responses.
A larger context window enables the model to analyse lengthy reports, extensive codebases, research papers and other complex documents without losing context. The model debuted on Arena.AI, a crowdsourced benchmarking platform where users compare AI systems across various tasks.
Following its debut, Qwen3.8-Max became the highest-ranked Chinese text model on the platform, though it still trails Anthropic's Claude Fable 5 and three Claude Opus variants. It also secured second position globally for image understanding and other visual reasoning tasks, behind only a Claude Fable 5 variant.
Designed for efficiency as well as performance
Despite its enormous size, Alibaba says Qwen3.8-Max has been designed to operate efficiently. The model uses a mixture-of-experts architecture, an approach that activates only the parts of the model required for a specific task instead of running the entire system.
Although the model contains 2.4 trillion parameters, only around 95 billion are active during any single request, helping reduce computing costs while improving response speeds.
Alibaba also revealed that the model completed an internal software engineering project over a period of 16 days, highlighting its potential for coding, enterprise automation and complex technical workflows.
Qwen3.8-Max is currently available in preview through Alibaba's Token Plan, Qoder and QoderWork platforms, with broader availability via Alibaba Cloud's Model Studio expected as competition in the global AI market continues to intensify.