The dedicated Tongyi team at Aliyun has proudly introduced Qwen3Guard, marking a significant milestone as the inaugural security protection guardrail model within the esteemed Qwen family. Built upon the robust Qwen3 architecture, this innovative model has undergone meticulous fine-tuning specifically tailored for security classification tasks. Its primary objective is to furnish comprehensive security assurances during AI interactions. Qwen3Guard excels in its ability to accurately pinpoint potential risks, while simultaneously generating detailed risk levels and precise classification labels.

Notably, Qwen3Guard has showcased exceptional prowess on leading security evaluation benchmarks, solidifying its position as a frontrunner in the field. Moreover, its versatility extends to multilingual scenarios, catering to a global audience. The model is available in two specialized professional versions: Qwen3Guard-Gen and Qwen3Guard-Stream. Each variant comes in three distinct parameter sizes: 0.6B, 4B, and 8B, ensuring flexibility and scalability. Furthermore, it boasts extensive language support, encompassing 119 languages and dialects, thereby facilitating seamless communication across diverse linguistic landscapes.