The Chinese tech giant Alibaba has launched a new line of AI models called Qwen2.5-VL, which are capable of analyzing text, images, and video, as well as controlling devices on PCs and smartphones. This was reported by Reuters.
The company stated that the model surpasses not only GPT-4o (OpenAI), Claude 3.5 Sonnet (Anthropic), and Gemini 2.0 Flash (Google), but also the Chinese DeepSeek, which recently became a sensation, led global downloads, and caused a drop in the stocks of American tech companies.
What Qwen2.5-VL can do:
- recognize all types of documents, including handwritten text, tables, graphs, chemical formulas, and handwritten notes;
- recognize, search for, and count objects in various formats, including absolute coordinates and JSON;
- analyze videos lasting several hours, highlighting key events in just a few seconds;
- make decisions and control devices, enhancing the model's functionality as a digital assistant.
Alibaba showcased a video where Qwen2.5-VL books a ticket in the Booking.com app on Android and manages programs in Linux, although currently, the AI model only handles basic commands.
https://twitter.com/_philschmid/status/1883960019030524346
Like other Chinese AIs, Qwen2.5-VL has built-in filters. When asked about politically sensitive topics, such as the mistakes of China's leader Xi Jinping, the chatbot simply refuses to respond. This is due to requirements from Chinese regulators that compel developers to adhere to "core socialist values."
Experts believe that the launch of Qwen2.5-VL is a response to the rapid growth of DeepSeek.
In 2024, DeepSeek-V2 triggered a price war in China by offering services at a record low price of 1 yuan ($0.14) for 1 million tokens. This forced Alibaba, Baidu, and Tencent to reduce the prices of their models by 97%.
Nevertheless, DeepSeek's founder, Liang Wenfeng, claims that price wars do not interest him: "Our goal is to achieve AGI (Artificial General Intelligence - note by Spot), which will surpass human intelligence."
Previously, Spot reported that DeepSeek topped the list of free applications in the Uzbek App Store.