Qwen3-Omni-Flash - Alibaba Tongyi's Omni-Modal Large Model

What is Qwen3-Omni-Flash

Qwen3-Omni-Flash (Qwen3-Omni-Flash-2025-12-01) is an omni-modal large model launched by Alibaba's Qwen team. The model can seamlessly process multiple input forms such as text, images, audio, and video, and generate high-quality text and natural speech output in real time. The model is based onQwen3-Omni and has been comprehensively upgraded in areas such as audio-video interaction, system prompt control, and multilingual interaction. The model has stronger instruction-following capabilities and more natural and fluent speech performance, and is committed to providing users with an AI interaction experience of "sound, form, and meaning in harmony, with intelligence following every command." It is currently a frontier product in the field of omni-modal AI.

Main Features of Qwen3-Omni-Flash

• Multimodal input and output: Supports multiple input forms such as text, images, audio, and video, and generates high-quality text and natural speech output in real time.
• Audio-video interaction: The model significantly improves the understanding and execution of audio-video instructions, enhances the stability and coherence of multi-turn dialogue, and delivers more natural and fluent speech performance.
• System Prompt control: Fully opens custom permissions, allowing users to finely control model behavior, set persona style, colloquial preferences, and response length, etc.
• Multilingual support: Supports 119 text languages, 19 speech recognition languages, and 10 speech synthesis languages, ensuring accurate interaction in cross-language scenarios.

Performance of Qwen3-Omni-Flash

• Stronger text understanding and generation: Achieves major improvements in tasks such as logical reasoning (ZebraLogic +5.6), code generation (LiveCodeBench-v6 +9.3, MultiPL-E +2.7), and comprehensive writing (WritingBench +2.2), bringing complex instruction-following capabilities to a new level.
• More accurate speech understanding: Character error rate in speech recognition (Fleurs-zh) is significantly reduced, the score in speech dialogue evaluation (VoiceBench) improves by 3.2 points, and speech understanding capability is enhanced.
• More natural speech generation: Multilingual speech synthesis quality is comprehensively improved, especially in Chinese and multiple languages, with rhythm, speed, and pauses closer to real human dialogue.
• Deeper image understanding: Achieves leaps in multidisciplinary visual question answering (MMMU +4.7, MMMU_pro +4.8) and mathematical visual reasoning (Mathvision_full +2.2), enabling more accurate "understanding" of image content and in-depth analysis.
• More coherent video understanding: Video semantic understanding capability (MLVU +1.6) continues to be optimized, and combined with enhanced audio-video synchronization capability, provides a solid foundation for real-time video dialogue.

Qwen3-Omni-Flash Project Address

• Project official website：https://qwen.ai/blog?id=qwen3-omni-flash-20251201

How to use Qwen3-Omni-Flash

• QwenChat website: VisitQwen Chatthe official website to interact directly with the model and experience text, speech, and image processing functions.
• Alibaba Cloud Model Studio platform: VisitAlibaba Cloud Model Studioofficial website and search for "qwen3-omni-flash-realtime-2025-12-01" to integrate the model into applications via API calls and implement customized functions.