[Alibaba Tongyi Qianwen Releases Qwen3.8-Omni-Flash, Reducing Video Input Costs by Approximately 89% Compared to Previous Generation] Alibaba Tongyi Qianwen has launched its first all-modal model, Qwen3.8-Omni-Flash, designed around intelligent agent capabilities. This model integrates native audio-video understanding, reasoning, and tool invocation into a single framework, enabling a complete process of content comprehension, task planning, tool utilization, and result delivery. The model supports a 1 million token context, allowing it to actively locate key segments within long videos. In the OmniVideoBench test, its intelligent agent perception mode reduced token consumption by 51.8% compared to static understanding, while video input costs decreased by approximately 89% compared to Qwen3.5-Omni-Plus.