qwenlm/qwen3-omni is an omnimodal foundation model project developed by the Qwen team at Alibaba Cloud. According to the repository description, it understands text, audio, images and video and generates real-time speech. This entry is a project introduction, not the release of a new version. FACTS records no latest release, so no version changes can be confirmed.
The project presents an end-to-end model that handles several input formats in one model and also generates speech. End-to-end means that input interpretation and response generation are handled within a single model system. The README summary describes native handling of multilingual input and streaming responses in text and natural speech. It also introduces audio and video interaction, system prompt control, and multimodal use cases. The repository license is Apache-2.0, and the language is listed as Jupyter Notebook. The star count recorded in FACTS is 4062.
The described features should nevertheless be distinguished from actual performance. FACTS contains no benchmark figures showing response latency or recognition accuracy, and records no latest release. Real-time speech generation and low-latency interaction should therefore be read as the project's own feature claims, and the record alone cannot indicate performance level or readiness for deployment.