AlibabaQwen Guard
Qwen3Guard-Gen-4B
Qwen3Guard-Gen treats moderation as a generative instruction task. It is trained from Qwen3 and supports 119 languages and dialects.
Practical use
Practical guidance. Check mixed-language content and quoted harmful text separately. A moderation pipeline needs to distinguish a prohibited request from reporting, criticism or a safety explanation.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | Downloadable |
| Model identifiers | Qwen/Qwen3Guard-Gen-4B |
| Training family | The Guard series uses 1.19 million labelled prompts and responses.Qwen3Guard-Gen-4B model card |
| Interface | Generative safety classification.Qwen3Guard-Gen-4B model card |
| Licence | Apache 2.0.Qwen3Guard-Gen-4B model card |
Related models
- Qwen3Guard-Gen-0.6B
Alibaba · Qwen Guard
A 600-million-parameter Qwen3 moderation model for complete text in 119 languages and dialects.
Launch date not confirmedDownloadable - Qwen3Guard-Gen-8B
Alibaba · Qwen Guard
Eight-billion-parameter Qwen3-based model that treats text moderation as an instruction task.
Launch date not confirmedDownloadable - Qwen3Guard-Stream-0.6B
Alibaba · Qwen Guard
A 600-million-parameter streaming moderation model with token-level classification.
Launch date not confirmedDownloadable