Alibaba open-sources Qwen3.8-27B model on August 15
Qwen3.8-27B runs locally on consumer GPUs with multimodal support
Alibaba is set to open-source its Qwen3.8-27B large language model at midnight on August 15, positioning it as the successor to the popular Qwen3.6-27B. This new model distills advanced capabilities from the higher-tier Qwen3.8 Max—including coding, professional work, and scientific research—into a lighter, locally executable format. Designed to run on consumer-grade hardware like high-end GPUs and PCs with Ryzen AI Max processors, Qwen3.8-27B introduces native multimodal support for images and video, comprehensive STEM diagram and document analysis, and customizable reasoning depth. It also features enhanced agentic capabilities, enabling autonomous planning and environmental feedback for complex multi-step tasks.
This release marks a strategic pivot in Alibaba's open-model approach, following a pattern of alternating between open and closed releases (Qwen3.5/3.6 open, Qwen3.7 closed, Qwen3.8 open again). Notably, Alibaba is also introducing a revenue-sharing model for large commercial users of its premium Qwen3.8 Max, requiring them to share a portion of revenue—a move that mirrors Moonshot AI's Kimi K3 License. While specifics of the revenue-sharing percentage are still under negotiation, this could set a precedent for monetizing open models in the Chinese AI ecosystem.
- Qwen3.8-27B will launch as an open-weight model on August 15 via ModelScope and Hugging Face
- Supports local deployment on consumer GPUs and Ryzen AI Max PCs with multimodal and agentic features
- Alibaba plans revenue-sharing for commercial users of Qwen3.8 Max, following Moonshot AI's Kimi K3 License model
Why It Matters
Accelerates local AI adoption with lightweight models while introducing new monetization for open-source ecosystems.