Alibaba's new Qwen3.8-Omni-Flash model can now process text, images, audio, and video simultaneously. Its main achievement is the ability to handle up to 1 million tokens in a single request. This opens doors for analyzing very long video or audio recordings, creating detailed summaries of multi-hour meetings, or conducting in-depth research where every piece of information is crucial.
The manufacturer claims that compared to the previous version, Qwen3.5-Omni-Plus, the new model has improved its performance in 30 different tests by over 26%. Progress is particularly noticeable in tasks related to audio and video analysis, code writing, and handling large context volumes. This means the model is now better at tasks requiring an understanding of complex relationships within data, whether it's a sequence of frames in a video or a long piece of program code.
For developers, this signifies new opportunities in creating applications for multimedia analysis, automatic subtitle generation for long videos, or even assisting in writing more complex software. The reduction in API request costs to 0.8 yuan per million tokens makes such advanced capabilities more accessible.
Now that AI can "watch" and "listen" as long as a human, or even longer, we can expect the emergence of tools that previously seemed like science fiction. For example, automatic analysis of an entire movie for specific scenes or themes, or instantly obtaining a complete transcript and key takeaways from a multi-hour conference.