According to CoinMeta, the Qwen3.8-Omni-Flash model was launched by Alibaba Qianwen and supports text, images, audio, video, and up to 1 million words of context, significantly enhancing the agent capabilities of the model. This model is capable of extracting information from videos, planning tasks, and invoking external tools for editing, dubbing, and rendering. The official team demonstrated its applications in scenarios such as long meeting processing, short drama translation, and movie commentary. Based on omnivideobench data, the model's accuracy has increased from 63.4% to 67.8%, while token consumption has been reduced by 45.7%. In addition, the cost of audio input has decreased by over 98%, and the cost of audio and video input has decreased by over 93%. Currently, the model weights have not been made public; only API is available.