Qwen3.8-Max

Provides precise safety detection for both prompts and responses, complete with risk levels and categorized classifications for accurate moderation. It achieves state-of-the-art performance on major safety benchmarks across English, Chinese, and multilingual environments.

Extends Qwen-Image’s unique text rendering capabilities to image editing tasks, enabling precise text editing. It simultaneously feeds the input image into Qwen2.5-VL for visual semantic control and a VAE Encoder for visual appearance control, achieving both semantic and appearance editing.

Excels at complex text rendering within images, including multi-line layouts, paragraph-level semantics, and fine-grained details. This 20B MMDiT image foundation model supports alphabetic languages.

Enables high-quality translation across 92 major official languages and prominent dialects, covering over 95% of the global population. This model builds upon Qwen3 and leverages reinforcement learning to comprehensively enhance multilingual understanding, translation accuracy, and linguistic fluency.

Proposes and utilizes the Group Sequence Policy Optimization (GSPO) algorithm to enable successful and stable scaling of Reinforcement Learning for language models. This addresses severe instability issues and model collapse observed in existing RL algorithms during long training.