The new Qwen3.8-Omni-Flash AI from Alibaba features native audio-video understanding and tool use to automate tasks such as vlog editing and movie recapping. The model supports a 1 million token context window and showed an average increase of 19.5 points in agent performance across WildClawBench-MM and UniClawBench benchmarks. Its video input costs are approximately 89% lower than those of the previous Qwen3.5-Omni-Plus version.

Alibaba open sourced Qwen-MM-Plugins and the Qwen-Live Harness to facilitate the development of applications using the Omni ecosystem. Qwen3.8-Omni-Flash reduces token usage by 51.8% on OmniVideoBench compared to static understanding when locating key moments in long videos. The model approaches Gemini 3.8 Flash in audio-video capabilities and allows users to plan and execute tasks through agentic perception.