Alibaba Introduces Qwen3.8-Max, Its Most Powerful AI Model Yet Amid Growing Industry Rivalry

Chinese technology giant Alibaba has introduced Qwen3.8-Max, describing it as the company's biggest and most advanced artificial intelligence model to date. The announcement comes just weeks after rival Moonshot AI launched its Kimi K3 model, highlighting the rapidly intensifying competition among China's leading AI developers.

Qwen3.8-Max features 2.4 trillion parameters, placing it close to Moonshot AI's 2.8 trillion-parameter Kimi K3. Parameters are the internal values an AI model learns during training, enabling it to recognize patterns, generate responses, and perform complex tasks. While a larger parameter count does not automatically guarantee superior performance, it remains an important benchmark for measuring the scale and capability of modern AI systems.

Chinese AI companies have increasingly emphasized publishing parameter counts to attract developers, particularly because many of their models are released as open-weight systems. This allows developers to download and adapt the models for their own applications. In contrast, major U.S. AI firms such as OpenAI, Anthropic, and Google keep the architecture and parameter counts of their flagship closed-source models confidential.

Alibaba unveiled Qwen3.8-Max on Arena.AI, a crowdsourced platform that compares AI model performance. The model quickly became the highest-ranked Chinese text-based AI model on the platform, although it still trails Anthropic's Claude Fable 5 and several Claude Opus variants in overall text performance. The model performed even more impressively in visual AI benchmarks. On Arena.AI's leaderboard for image and multimodal analysis, Qwen3.8-Max secured the second position globally, ranking behind only one version of Claude Fable 5.

Like Moonshot AI's Kimi K3, Alibaba's latest model supports text, image, and video understanding while offering a context window of up to one million tokens. A larger token limit enables the AI to process massive amounts of information in a single session, including lengthy legal documents, extensive software code, or hundreds of pages of text without losing context.

Alibaba also said Qwen3.8-Max is built using a Mixture-of-Experts (MoE) architecture. Rather than activating the entire model for every request, the system selectively engages specialized components, meaning only 95 billion parameters are active during inference. This approach helps lower computing costs while improving response speed.

According to the company, the model recently completed a complex software engineering project in just 16 days, demonstrating its ability to handle long-duration development tasks. Alibaba plans to make Qwen3.8-Max publicly available next week through its Alibaba Cloud Model Studio platform, giving developers and enterprises access to its latest flagship AI system.