Alibaba Open Sources Qwen3.8 Series: 27B Multimodal Model Outperforms Predecessor

Alibaba Releases Qwen3.8 Series as Open Source

On August 15, Alibaba announced the open-source release of its Qwen3.8 model series. The models are now freely accessible for download, deployment, and use by developers, research institutions, and companies worldwide.

Spotlight on Qwen3.8-27B: A Multimodal Leap

The flagship model in this release is Qwen3.8-27B. It is a natively multimodal dense model built with 27 billion parameters. Despite its efficient size, its overall performance benchmark exceeds that of the previous Qwen3.7-Plus model.

Technical Edge and Broader Implications

The native multimodal design enables the model to process and integrate different data types, such as text and images, more seamlessly.

• Parameter Efficiency: With 27B parameters, it balances strong capability with reduced computational demands, lowering the barrier for practical deployment and fine-tuning.
• Performance Gain: Surpassing the prior Plus version indicates significant improvements in model architecture and training efficiency.

By open-sourcing these models, Alibaba aims to foster wider innovation and adoption across the AI ecosystem, from academic research to commercial applications.