Alibaba is preparing to release a new Qwen 3.5 model

In the world of artificial intelligence, major changes are coming: according to 36氪, Alibaba is preparing to release a new version of its foundational model Qwen, named Qwen 3.5. Evidence of this comes from the appearance of a request to include Qwen 3.5 in the Transformers library on the HuggingFace platform, the world's largest open-source AI community. This move signals an imminent release and generates significant interest in the industry.

What is Qwen and why does this matter? Qwen is a family of large language models (LLM) developed by Alibaba Group. These models, like other LLMs, are capable of generating text, translating languages, writing various types of creative content, and answering your questions in an informative manner. They are cornerstones of many modern AI applications, from chatbots to automatic content generation systems. The upgrade to version 3.5 means significant improvements in performance, efficiency, and functionality.

According to available information, Qwen 3.5 uses a new hybrid attention mechanism (mixed attention mechanism). This could lead to increased efficiency in information processing and improved quality of generated text.

But what is even more interesting, Qwen 3.5 is likely a Vision-Language Model (VLM), inherently capable of understanding visual information. This means the model will be able to process images and video as well as text, opening new possibilities for creating multimodal AI applications.

Developers have also discovered that Alibaba plans to open-source at least two versions of Qwen 3.5: a dense model with 2 billion parameters and a Mixture of Experts (MoE) model with 35 to 343 billion parameters. Opening the source code will allow researchers and developers worldwide to study, adapt, and improve the model, which will contribute to the development of AI technologies.

The release of Qwen 3.5 has serious implications for the artificial intelligence industry. First, it strengthens Alibaba's position as one of the leaders in LLM development. Second, support for visual understanding opens new opportunities for creating AI applications that combine text and images. Third, opening the source code of Qwen 3.5 models will promote the development of the AI community and accelerate innovation in this field. For users, this means the emergence of more powerful and versatile AI tools that will be able to solve a wide range of tasks, from automatic content creation to image and video analysis.

In conclusion, the release of Qwen 3.5 is an important step forward in the development of large language models. The new architecture, support for visual understanding, and opening of the source code make this model a promising tool for researchers, developers, and users. Alibaba continues to demonstrate its commitment to innovation in artificial intelligence, and Qwen 3.5 is clear evidence of this. Now we await the official release to see how Qwen 3.5 performs in practice.