Alibaba launches AI model that understands audio and video

The Chinese tech company Alibaba has launched a new AI model called Qwen3.8-Omni-Flash. The model can understand text, images, audio and video at the same time, and it can use digital tools on its own to solve tasks, for example searching the web.

One of the big pieces of news is that the model can analyze long videos and make summaries of them. It can also respond in real time, which makes it well suited for tasks where you need a quick answer while you're talking or showing something.

The model has a so-called context window of up to 1 million tokens. In short, tokens are the pieces of text and data an AI model works with at a time. A large context window means the model can remember and work with large amounts of information in one go, for example an entire long video or many documents at once.

According to Alibaba, the new model performs significantly better than the previous version, Qwen3.5-Omni-Plus, with an improvement of more than 25 percent across a range of tests. The price for using audio in the model has also been cut sharply, by more than 98 percent, compared to the old model.

Qwen3.8-Omni-Flash is so far available only through Alibaba's own cloud platform, meaning as paid access for developers and companies. No so-called open weights have been released, which has otherwise been a hallmark of several earlier Qwen models, so this time you can't download and run the model on your own server.

Ordinary Danish users typically won't encounter the model directly, but in practice through apps and services built on top of Alibaba's technology, for example tools for automatic meeting summaries or video analysis.

What it means for you

You probably won't use the model directly, but you'll encounter it through apps and services that use it for things like meeting summaries or video analysis. If you're a developer or a company, you can build solutions with the model yourself via Alibaba's cloud platform. Be aware that you can't download and run the model yourself, as it's only available as paid access.