Alibaba's Qwen team has unveiled Qwen3.8-Max, their most capable model to date, featuring a colossal 2.4-trillion-parameter Mixture-of-Experts (MoE) architecture. This multimodal flagship processes text, images, and video inputs to generate text outputs. Alongside, the more accessible Qwen3.8-27B is also being released. Both models will soon offer open weights, complementing the immediate availability of Qwen3.8-Max via a hosted API that boasts OpenAI and DashScope compatibility for seamless integration.
Qwen3.8-Max leverages its MoE design to handle complex tasks, supported by an impressive 1-million-token context window for deep data understanding. Developers gain access to a robust feature set, including function calling, structured outputs, batch processing, and fine-tuning, plus five integrated tools like a code interpreter and web search. While Qwen3.8-Max's open weights demand multi-node datacenter infrastructure, Qwen3.8-27B is optimized for standard on-premise GPUs, broadening deployment options.
This release significantly empowers developers and researchers with unparalleled scale and multimodal processing. Its vast context window and specialized tools are ideal for creating advanced applications such as repository-scale coding agents, comprehensive long-document knowledge bases, and sophisticated multi-step research assistants. By addressing key needs in software engineering, legal, finance, media, e-commerce, and design, Qwen3.8-Max is poised to accelerate innovation across diverse industries.