AI News (2026/3/3): Alibaba Qwen Open-Sources Four Qwen3.5 Small-Scale Models
Executive Summary:
Alibaba Qwen has announced the open-sourcing of four Qwen3.5 small-scale models, namely Qwen3.5-0.8B, 2B, 4B, and 9B. These models use native multimodal training and the latest architecture, designed to meet lightweight AI demands from edge devices to servers. The 0.8B and 2B models are suitable for mobile devices and IoT edge deployment, the 4B model is positioned as a lightweight Agent base, and the 9B model matches the performance of gpt-oss-120B.
Alibaba Qwen Open-Sources Four Qwen3.5 Small-Scale Models
Details
Alibaba Qwen, in the latest research from the Tongyi lab, has announced the open-sourcing of four Qwen3.5 small-scale models. These models, with parameter sizes of 0.8B, 2B, 4B, and 9B, utilize native multimodal training and the latest architectural design. Through this series of models, Alibaba Qwen aims to meet diverse lightweight AI demands from edge devices to servers.
Key Points
【Small-Scale Models】: The parameter sizes of the Qwen3.5 series models are 0.8B, 2B, 4B, and 9B. These models maintain high performance while significantly reducing the demand for computational resources, enabling them to run efficiently on various hardware platforms.
【Multimodal Training】: All Qwen3.5 models are trained using native multimodal training technology, capable of handling multiple data types such as text and images. This training method not only enhances the model's generalization ability but also makes it more flexible in practical applications.
【Deployment Range】: The Qwen3.5-0.8B and 2B models are particularly suitable for mobile devices and IoT edge deployment, enabling real-time inference and low-latency responses. The 4B model is positioned as a lightweight Agent base, ideal for scenarios requiring higher computational power but with limited resources. The 9B model matches the performance of gpt-oss-120B, suitable for high-performance server applications.
AI-ALL In-Depth Commentary
The open-sourcing of the Qwen3.5 small-scale model series by Alibaba Qwen not only enriches the existing AI model ecosystem but also provides developers with more options. Particularly, the introduction of the 0.8B and 2B models makes AI technology more viable for mobile devices and IoT edge computing. The 4B model, as a lightweight Agent base, is expected to drive the development of more intelligent agent applications. The performance breakthrough of the 9B model offers new solutions for high-performance server applications.
The open-sourcing of this series of models will further promote the popularization and innovation of AI technology. Developers can quickly build and optimize their applications using these models without having to train large-scale models from scratch. Additionally, this move by Alibaba Qwen demonstrates its commitment and determination to promoting the open sharing of AI technology.