Qwen has rolled out several new AI models, emphasizing smaller sizes with high performance. The Qwen3-VL-30B-A3B-Instruct, for instance, operates with just 3 billion active parameters but competes effectively against larger models like GPT-5-Mini and Claude4-Sonnet in various tasks, including STEM and video analysis. Notably, there's an FP8 version available to enhance speed for multimodal applications. Users can access these models through different platforms, including Qwen Chat and GitHub. Another significant release is Qwen3-Omni, which combines text, image, audio, and video processing in a single model without compromising quality. It achieves state-of-the-art performance on 22 out of 36 audio and audiovisual benchmarks, showcasing its capabilities in understanding and generating content across multiple formats. Developers can leverage open-sourced versions of this model, including those tailored for instruction-following and creative tasks. Qwen is also pushing the boundaries of efficiency with the Qwen3-Next-80B-A3B model. This model uses 80 billion parameters but only activates 3 billion at a time, making it cheaper and faster than previous versions. It features a hybrid architecture designed for speed and recall, and is said to outperform several competitors in reasoning tasks. Lastly, the introduction of Qwen-Image-Edit focuses on precise bilingual text editing and semantic manipulation, enhancing creative and practical applications.