Qwen3.8-Omni-Flash integrates native audio-video reasoning and tool use to automate tasks like vlog editing and video translation. The model features a 1 million token context window and an average performance increase of 19.5 points in agent capabilities across WildClawBench-MM and UniClawBench benchmarks. Video input costs are 89% lower than the previous Qwen3.5-Omni-Plus version.

Alibaba is open-sourcing the Qwen-MM-Plugins and Qwen-Live Harness to support developer application building. The model uses 51.8% fewer tokens than static understanding for locating key moments in long videos on OmniVideoBench and approaches the audio-video capabilities of Google's Gemini 3.8 Flash.