The most vivid star of the Chinese industry of artificial intelligence in the last days has become the DeepSeek laboratory, but the technological giants are not sitting in terms of: the Alibaba QWEN unit presented the QWEN2.5-VL family capable of managing the PC and a smartphone, as the virtual assistant Openai Operator does.
Image source: alibababacloud.com
The most powerful model in the QWEN2.5-VL family surpassed the largest American projects, including Openai GPT-4O, Anthropic Claude 3.5 Sonnet and Google Gemini 2.0 Flash in a number of tests, including “understanding” video, solving mathematical problems, analysis of documents and answers For questions, the developers say. You can test this model in the Alibaba Qwen Chat application, it is available from the Huging Face platform. She analyzes diagrams and graphs, extracts data from accounting documents, studies many hours of video, and also recognizes fragments of films and series – perhaps her training was conducted using copyrights protected by copyright. Like other Chinese models, she refuses to comment on Beijing’s policy.
Image source: X.com/_philschmid
One of the most interesting features of QWEN2.5-VL is its ability to manage programs on PC and mobile devices. In one example, the model launched the Android application and booked flights. In another example, she was entrusted with the Office of the PC for Linux, but she was able to perform only basic actions, in particular, she switched the tabs in the browser. The younger versions of QWEN2.5-VL-3B and QWEN2.5-VL-7B are available in an open license without restrictions; The flagship QWEN2.5-VL-72B requires that the owners of platform owners with more than 100 million users receive permission from Alibaba QWEN before commercial deployment of the model.