Qwen3.8-Omni: Towards Native Omni-Modal Agents

Qwen3.8-Omni-Flash is presented as a natively multimodal agentic model built for real-world multimodal productivity. Whereas earlier omni models concentrated mainly on perception and interaction, this release substantially improves multimodal understanding and reasoning as well as performance on long-horizon agentic tasks.

Those gains come from a native multimodal co-training strategy that preserves strong text-domain capabilities while transferring agentic capabilities from text to audio and video tasks. Architecturally, the model inherits the sparse mixture-of-experts (MoE) design of Qwen3.8-Next and extends the context window to one million tokens, enabling long-context multimodal reasoning and long-horizon planning.

These capabilities let the model serve in production workflows either as a primary agent or as a specialized sub-agent, with cited applications including video editing, long-form audio and video translation, music-conditioned music video or movie generation, and video-based note or omni-skill creation.

To address the absence of native audio and video support in existing agent harnesses, the team released Qwen-MM-Plugins, a lightweight open-source plugin framework for multimodal productivity. It also frames real-time multimodal interaction as a system-level challenge requiring orchestration of context and memory management, tool use, and sub-agent delegation, and accordingly releases Qwen-Live-Harness, an open-source framework for building responsive, real-time multimodal agents on top of Qwen3.8-Omni-Flash.

Extensive evaluations show strong performance across multimodal understanding, reasoning, long-horizon agentic execution, and video productivity tasks. The team positions these results plus the accompanying open-source tools as making Qwen3.8-Omni-Flash a practical foundation for deploying natively multimodal agents in both research and production.