A new artificial intelligence model from China’s Alibaba has quickly become the most popular system on Hugging Face, the world’s largest open-source AI platform.
The model, called Qwen3-Omni, can understand and generate text, images, audio and video – all in one system. This makes it what’s known as a multimodal model. In simple terms, multimodal AI can work across different types of input and output, rather than being limited to just text or just images.
For example, you could show Qwen3-Omni a video, ask it to describe what is happening, and then request a summary in another language – all without switching tools.
Within days of its release, Qwen3-Omni climbed to the top of Hugging Face’s trending list, beating out well-known Western systems. Another Alibaba product, Qwen-Image-Edit, took second place.
Why Qwen Stands Out
One of Qwen’s biggest advantages is that it’s open-source. This means the code and model weights are free to download and adapt under a permissive Apache 2.0 licence. Researchers, start-ups and developers can build on it without the heavy costs or restrictions that come with closed models.
By contrast, OpenAI’s GPT models and Google’s Gemini remain largely closed, limiting how much developers can experiment with them. This openness has helped Qwen gain a rapid following worldwide.
In benchmark tests, Alibaba reports that Qwen3-Omni outperforms GPT-4o and Gemini 2.5-Flash in areas like recognising speech, understanding video, and interpreting images.
From Text to Video – How It’s Evolving
Qwen hasn’t appeared overnight. The first version launched in April, focused mainly on text. Since then, the family has grown to include models for coding, reasoning, image editing and video analysis.
Recent upgrades include:
- Qwen3-VL – a vision-language model able to analyse hours of video, interact with apps, and even convert sketches into computer code. It can also handle 32 languages and huge amounts of data, such as entire textbooks or full-length films.
- Qwen3-Next – a new architecture designed for “ultra-long” context, allowing it to process information faster and remember more. This is seen as a stepping stone towards the next big version, Qwen3.5.
These improvements show how quickly Alibaba is moving to challenge US rivals.
Hugging Face Popularity
The open-source community has responded enthusiastically. This week, Alibaba’s models held five of the top ten trending spots on Hugging Face. Other Chinese firms, such as Tencent and DeepSeek, filled four more – leaving IBM as the only Western company on the list.
In total, Alibaba has released more than 300 open-source models, which have already inspired over 170,000 spin-offs by developers around the world.
The Bigger Picture
The rise of Qwen reflects a wider shift in AI. In the US, leading models are becoming increasingly closed and controlled by large tech firms. Meanwhile, Chinese companies are leaning into openness, making powerful tools widely available for research and business.
Some experts warn this could see the balance of AI leadership tilt towards China. For now, though, what’s clear is that Qwen has become a major force in the race to build the next generation of artificial intelligence.
