The Chinese Ki Laboratory Deepseek could attract a lot of the attention of the Tech industry this week. But among the best domestic competitors, Alibaba, shouldn’t be idle.
Alibabas Qwen team on Monday released A brand new family of AI models, QWEN2.5-VL, which might perform numerous text and image evaluation tasks. The models can analyze files, understand videos and count objects in images and control a PC – much like the recently launched operator of the Openai model.
According to the benchmarking of the QWen team, the most effective QWen2.5 VL model Openas GPT-4O, the Claude 3.5 sonet from Anthropic and Google's Gemini 2.0 Flash in numerous video understanding, mathematics, document evaluation and questions.
QWen2.5-VL that may be tested in Alibaba's Qwen chat App and too download The face hugs from the AI DEV platform, can analyze diagrams and graphics, extract data from scans of invoices and forms and “understand” videos several times, based on the QWen team. Qwen2.5-VL can ie Recognize “Ips from Film and TV Series, in addition to a wide selection of products,” by team – This concludes that the models can have been partially trained in copyrighted works.
QWEN2.5-VL, which was developed by a Chinese company, has certain restrictions on the topics that he explains within the QWen chat. When I asked the most important and capable Qwen2.5 VL model, Qwen2.5-VL-72B, talking about “XI Jinping errors”, Qwen Chat threw an error message.
China's Internet regulator Benchmarks Many models developed within the country to make sure their answers “embody the core socialist”. Many Chinese AI systems refuse to react to topics that would increase the anger of the supervisory authorities similar to the autonomy of Taiwan.
One of the more interesting functions of QWEN2.5-VL is the power to make use of software to interact with interacting on PCs in addition to on mobile devices. A video published by Philipp Schmid on X, a technical advantage at Hugging Face, showed Qwen2.5-VL, which began the Booking.com app for Android and booked a flight from Chongqing to Beijing.
Don't miss @Alibaba_qwen 2.5 VL! Despite all of the Deepseek hype, Qwen has just fallen the most effective open multimodal! QWEN 2.5 VL is an vision model that may control your computer, much like that @Openai Operator, structured information from diagrams and more extract !!
Tl;
3️⃣ … pic.twitter.com/geegvdl0ti– Philipp Schmid (@_philschmid) January 27, 2025
In the next video, a QWEN2.5 VL model controls apps on a Linux desktop also not to realize much beyond the switching of tabs. Perhaps characteristically, Qwens Benchmarking Qwen2.5 VL value on Osworld, a benchmark that tries to mimic an actual computer environment.
LMAO QWEN 2.5 VL can use computer use and take over the Openai board head! 🐐 pic.twitter.com/lwmecxznsu
– Vaibhav (VB) Srivastav (@Reach_vb) January 27, 2025
The two smaller, less sophisticated models within the QWEN2.5 VL series QWEN2.5 VL-3B and QWEN2.5-VL-7B can be found as a part of a permissible license. However, the flagship QWEN2.5-VL-72B is under Alibaba's custom license, during which corporations and developers with greater than 100 million lively users request the authorization of QWEN/Alibaba before the model is commercially provided.
