Qwen3Guard-Gen-8B

Qwen3Guard‑Gen‑8B is a generative safety model based on Qwen3, trained on 1.19 million safety-labeled prompts and responses. It classifies content into three levels: safe, controversial and unsafe. It supports more than 119 languages and can be integrated into moderation pipelines to classify both inputs and responses generated by language models.

• PROVIDER LOCATION: Non-EU
• PARAMETERS: 8B
• MAX CONTEXT: 32K token
• LICENSE: Apache 2.0
• FEATURES: Content moderation, Safety classification
• OUTPUT FORMATS: Raw_text (etichetta + categoria)

• Moderation of LLM inputs and outputs
• Content filtering for platforms
• Automated safety classification
• Compliance and policy enforcement

curl --request POST \
--url https://apiaruba.it/aiplatform/v1/chat/completions \
--header 'content-type: application/json' \
--header 'x-aruba-api-key: *****' \
--data '{
"model": "Qwen3Guard-Gen-8B",
"messages": [
{
"role": "system",
"content": "Sei un assistente conciso."
},
{
"role": "user",
"content": "Spiega cos'\''è un embedding."
}
],
"temperature": 0.7,
"max_tokens": 1000
}'

{
"id": "chatcmpl-ace0bc13-a6ec-4f68-bb68-c1da652b9446",
"created": 1787560998,
"model": "Qwen3Guard-Gen-8B",
"object": "chat.completion",
"choices": [
{
"finish_reason": "stop",
"index": 0,
"message": {
"content": "Safety: Safe\nCategories: None",
"role": "assistant",
"provider_specific_fields": {
"reasoning": null,
"refusal": null
}
},
"provider_specific_fields": {}
}
],
"usage": {
"completion_tokens": 8,
"prompt_tokens": 313,
"total_tokens": 321
}
}

• Hugging Face
• GitHub
• Paper