Alibaba introduced a new AI model Qwen 3.5, which is focused on use with visual agents. The model has 397 billion parameters, but it only uses 17 billion for operation. This has reduced the cost of use by 60 percent compared to the previous version and accelerated the processing of large volumes of data eightfold.
Qwen 3.5-Plus shows results comparable to Gemini 3 Pro and surpasses GPT-5.2 and Claude Opus 4.5 in tests. The model recognizes what is happening on the screen of a smartphone or computer and sees the interface elements of various programs.
The user can describe tasks in plain language, and Qwen 3.5 performs the necessary actions in applications, such as clicking, swiping, or entering text. This opens up new possibilities for automating work with various tools without additional training.
Alibaba has made Qwen 3.5 open for testing, so developers can already try its capabilities in practice. The model is available on GitHub and for testing , which will help quickly integrate it into their own projects and experiment with new usage scenarios.
