Qwen Ships Omni-Flash, Its First Agentic Omni-Modal Model

Alibaba's Qwen team has released Qwen3.8-Omni-Flash, which it describes as its first omni-modal model built specifically around agentic capability: native audio-video understanding, reasoning, and tool use combined in a single model rather than layered on top of a separate multimodal system.

Qwen says the model can jointly reason over what it sees and hears and orchestrate tools across long workflows, citing tasks like auto-editing vlogs, translating short videos, and turning full movies into recaps. The team reports the model approaches Gemini 3.8 Flash on audio-video capability and gains an average of 19.5 points in agent performance across the WildClawBench-MM and UniClawBench benchmarks, figures Qwen has not published independent third-party verification for in the material reviewed here.

With a 1-million-token context window, Qwen says the model actively explores long videos to locate key moments, using 51.8% fewer tokens than static understanding requires on the OmniVideoBench benchmark. The company also reports video input costs are about 89% lower than its prior Qwen3.5-Omni-Plus model.

Alongside the model, Qwen is open-sourcing two supporting projects: Qwen-MM-Plugins, available now, and a Qwen-Live Harness for real-time use that the company describes as coming soon.

The release continues a rapid pace of omni-modal launches from Chinese AI labs through 2026, aimed at making long-form audio-video agentic workflows cheaper to run. All performance and cost figures in this article come from Qwen's own announcement and have not been independently verified in the material reviewed here.