Alibaba Qwen Releases Multimodal Large Model Qwen3.5-Omni, Supporting Audio-Visual Understanding and Code Generation

Alibaba Qwen recently released the multimodal large model Qwen3.5-Omni, which can seamlessly understand text, images, audio and audio-visual inputs. It supports 256K long context and recognition of 113 languages, natively integrates WebSearch and complex tool calls, and can independently complete information queries and task execution.

The model has a naturally emerging Audio-Visual Vibe Coding capability, which can generate code based on video frames. It also supports semantic interruption, voice cloning and voice control, making conversations more like real human interactions. Additionally, it can generate fine-grained audio-visual captions with timestamps to help structure long video content.

Qwen3.5-Omni has achieved SOTA results in 215 tasks, outperforming Gemini3.1-Pro in audio and audio-visual analysis directions. Users can call its API through Alibaba Cloud's Bailian platform, with three sizes (Plus, Flash, Light) available to adapt to different scenario needs.