Qwen3.8-Omni-Flash: Native Omnimodal Agent Release
Qwen released Qwen3.8-Omni-Flash on September 18, 2026. This native omnimodal model supports text, image, audio, and video inputs with text and audio outputs…
Release Overview and Modalities
Qwen announced the release of Qwen3.8-Omni-Flash on September 18, 2026. The provider describes this model as a next-generation native omnimodal system. Its primary design objective is to strengthen agent capabilities within real-world productivity scenarios. The announcement states that the model aims to advance omnimodal systems from merely understanding content to actively planning tasks, calling tools, and completing creative work.
- Input modalities include text, image, audio, and video.
- Output modalities include text and audio.
- The model is identified as a native omnimodal architecture.
Agent Capabilities and Scope
The source material specifies that the model builds upon general agentic capabilities in coding, text-based knowledge work, and GUI interactions. The stated goal is to support workflows where the model plans actions and executes them through tool usage. The documentation does not provide specific benchmark scores, context window sizes, or maximum output token limits. It also does not confirm open weights availability, pricing structures, or specific deployment architectures. The term 'native omnimodal' indicates the model processes multiple data types inherently, but the source does not detail the internal mechanisms for how these modalities are integrated.
Evaluation Procedure for Engineers
To evaluate Qwen3.8-Omni-Flash, engineers should define specific productivity tasks that require multimodal input processing. First, prepare test datasets containing mixed media files, including images, audio clips, and video segments alongside text instructions. Second, measure the model's ability to parse these inputs by validating if it correctly identifies key information across different formats. Third, compare the model's output against expected results for task planning and tool calling sequences. Finally, validate the quality of the generated text and audio outputs against human-defined standards for accuracy and coherence. This process helps determine if the model meets specific operational requirements without assuming unverified performance metrics.
Primary sources
Frequently asked questions
What are the confirmed input and output modalities?
The source confirms that Qwen3.8-Omni-Flash accepts text, image, audio, and video as inputs. It produces text and audio as outputs. No other modalities are listed in the provided facts.
Is the model available with open weights?
The source records that open weights are not confirmed. It does not state whether local installation is possible or impossible, nor does it specify API-only access or hosting arrangements.