Qwen3.6-Plus

The Qwen family includes models like Qwen-Image and Qwen-Image-Edit, which are 20B MMDiT foundation models. They excel at complex text rendering, including multi-line layouts and paragraph-level semantics, and enable precise image editing, including semantic and appearance control by integrating with models like Qwen2.5-VL.

Qwen3Guard is the first safety guardrail model in the Qwen family, built upon Qwen3 foundation models. It is fine-tuned for safety classification, providing precise safety detection for both prompts and responses, complete with risk levels and categorized classifications for moderation across English, Chinese, and multilingual environments.

Qwen-MT (qwen-mt-turbo) offers high-quality translation across 92 major official languages and prominent dialects, covering over 95% of the global population. It is built upon Qwen3 and leverages trillions of multilingual and translation tokens, enhanced with reinforcement learning techniques for improved accuracy and fluency.

The Qwen team utilizes Reinforcement Learning (RL) and has proposed the Group Sequence Policy Optimization (GSPO) algorithm to enable stable and robust training dynamics for scaling language models, enhancing their deep reasoning and problem-solving capabilities.

The Qwen team embraces open-source development, having released various models in the Qwen series, including language models (Qwen), expert models (Qwen-Coder, Qwen-Math), and multimodal models (Qwen-VL, Qwen-Audio). Resources are available on platforms like GitHub, Hugging Face, and ModelScope.