Alibaba has unveiled Qwen3.8-Max, the most powerful model in its Qwen family, featuring 2.4 trillion parameters and support for a context window of up to 1 million tokens. The multimodal AI model is now available through APIs on Alibaba Cloud Model Studio, with model weights scheduled for release next week.
Built on the foundation of Qwen 3.5, Qwen3.8-Max uses a Sparse Mixture-of-Experts (MoE) architecture and hybrid attention mechanism, enabling it to activate only 95 billion parameters during inference despite its massive scale. This approach improves efficiency while maintaining advanced performance across coding, research, reasoning, and long-horizon tasks.
The model has achieved strong rankings across industry benchmarks, placing fifth in Text Arena, second in Vision Arena, and fourth in Frontend Code Arena. Alibaba highlighted the model’s ability to autonomously execute complex software engineering projects, including the creation of “oh-my-cli,” a self-evolving agent framework developed over a 16-day autonomous coding cycle and later released as open source on GitHub.
Beyond software development, Qwen3.8-Max demonstrated capabilities in scientific research, legal analysis, financial research, sports analytics, architecture, and other specialized domains. The company said the model can independently plan, execute, and refine tasks through closed-loop adaptive learning and system-level autonomous reasoning.
As a multimodal foundation model, Qwen3.8-Max also supports advanced visual intelligence. It can process large volumes of visual and textual content—including lengthy documents, television series, and livestreams—and convert them into searchable knowledge bases. The model can also generate educational animations, create interactive games, reconstruct web applications from screenshots, and transform 2D floor plans into 3D visualizations.
Alibaba said Qwen3.8-Max represents a significant step forward in autonomous AI agents, combining advanced reasoning, coding, and visual understanding capabilities to address increasingly complex real-world applications.
