Qwen Launches 30B Parameter Multi-Modal 'Omni' Model Amidst Suite of Four Major AI Releases

Alibaba Cloud's AI research team, Qwen, has unveiled a significant suite of four new artificial intelligence models, highlighted by the release of Qwen3-Omni, a new 30-billion parameter multi-modal model. This comprehensive launch, detailed in a blog post by AI expert Simon Willison, underscores Qwen's rapid advancement in the competitive AI landscape. The releases include advancements in multi-modal capabilities, speech synthesis, image editing, and more efficient large language models.

The flagship Qwen3-Omni model, available as open weights under an Apache 2.0 license, supports text, audio, and video input, alongside text and audio output, marking a substantial step forward in integrated AI functionalities. Willison described it as "today's most exciting announcement," noting its accessibility for running on powerful home devices due to its 70.5GB size. Users can interact with Qwen3-Omni via chat.qwen.ai, leveraging its voice and video chat features.

Accompanying Qwen3-Omni are the Qwen3-Next-80B-A3B-Instruct-FP8 and Qwen3-Next-80B-A3B-Thinking-FP8 models, which are official FP8 quantized versions of their Qwen3-Next series. These models offer enhanced efficiency, with the FP8 Instruct variant significantly reducing size from 163GB to 82.1GB, making them more manageable for deployment. Additionally, Qwen introduced Qwen3-TTS-Flash for multi-timbre, multi-lingual, and multi-dialect speech synthesis, accessible via API.

The fourth release is Qwen-Image-Edit-2509, an updated version of their image editing model that now supports multi-image inputs. This update reflects Qwen's commitment to regular iterations, with the company reportedly planning monthly enhancements for its image editing capabilities. The coordinated launch of these diverse models, spanning various AI domains, showcases Qwen's aggressive strategy in pushing the boundaries of open-source and proprietary AI development.