Model Details
Qwen3Guard-8B
Quick answer: Qwen3Guard-8B is Alibaba's 8B-parameter open-weights safety classifier, part of the Qwen3Guard family built for real-time (streaming) and generation-time content moderation. It scores highest among comparison models on the vendor-released Qwen3GuardTest benchmark and remains competitive across WildGuardTest and HarmBench.
At a Glance
Where it leads: WildGuardTest (Prompt) — narrowly the top score among Shieldstral's comparison set — and its own Qwen3GuardTest evaluation. Where it lags: Multimodal tasks (no vision support) and multilingual PolyGuard/RTP-LX benchmarks relative to purpose-built multilingual classifiers. Best for: Teams needing an open, mid-sized (8B) text safety classifier with both batch and streaming/real-time moderation modes.
What Qwen3Guard-8B Is
Qwen3Guard-8B is part of Alibaba's Qwen3Guard family, available in Gen (standard) and Stream (real-time, token-by-token) variants across multiple sizes. It's designed for prompt and response safety classification, trained on Alibaba's own safety taxonomy and evaluation methodology (Qwen3GuardTest).
Specifications
| Field | Value |
|---|---|
| Organization | Alibaba |
| Parameters | 8B |
| License | Apache 2.0 |
| Modality | Text only |
| Variants | Gen (standard), Stream (real-time) |
Pricing
Open weights, free to download.
Public Benchmark Scores
Scores from Mistral AI's Shieldstral model card, shown for context. Not Benchgen measurements. Figures reflect an average over strict/loose label mappings.
| Benchmark | Score |
|---|---|
| WildGuardTest (Prompt) | 88.2% |
| HarmBench (Prompt) | 99.3% |
| Qwen3GuardTest | 91.0% |
| Aegis v2 (Prompt) | 83.1% |
| XSTest (Harm) | 91.2% |
Related Models & Benchmarks
Frequently Asked Questions
What is Qwen3Guard-8B?
Alibaba's 8B open-weights safety classifier for prompt and response moderation, available in standard and real-time streaming variants.Is it multimodal?
No, it's text-only.Is it open source?
Yes, released under Apache 2.0.Scores sourced from Mistral AI's Shieldstral model card, shown for context. Last updated 2026-08-12.
Benchmark Leaderboards
This model isn’t on any benchmark leaderboard yet.