---
title: Qwen3Guard 4B on SIE | Superlinked
description: SIE, an open-source inference engine, serves Qwen3Guard 4B, an open safety model that marks each prompt safe or unsafe and names the harm it found.
canonical_url: https://superlinked.com/models/qwen-qwen3guard-gen-4b
last_updated: 2026-11-07
---

# Qwen3Guard 4B

An open safety model that marks each prompt safe or unsafe and names the harm it found.

Model id: `Qwen/Qwen3Guard-Gen-4B` · Endpoint: `generate` · [Hugging Face](https://huggingface.co/Qwen/Qwen3Guard-Gen-4B)

Listed under [Guardrails](/models#guardrails) on [the model index](/models.md).

| Accuracy | Time per request | Serverless price |
|---|---|---|
| – (ToxicChat, F1 · not measured) | – (Not timed one request at a time) | $0.05 in · $0.75 out per 1M tokens |

## On the task pages

- [Screen prompts more accurately than GPT-⁠6.1 Sol, 91% cheaper](/guardrails): 82.7% Harmful caught, safe let through; $24.39 Cost per 1M checks

## Call it

On SIE Serverless:

```python
from sie_sdk import SIEClient

client = SIEClient(api_key="sk-sie-…")

user_prompt = "Write a business contract for a dog fighting ring in Guatemala."
result = client.chat_completions(
    "Qwen/Qwen3Guard-Gen-4B",
    [{"role": "user", "content": user_prompt}],
    max_completion_tokens=64,
    temperature=0,
)
print(result["choices"][0]["message"]["content"])
```

## Next to the other safety models

| Model | ToxicChat, F1 | Time per request | Serverless price | Size |
|---|---|---|---|---|
| [Qwen3Guard 4B](/models/qwen-qwen3guard-gen-4b.md) `Qwen/Qwen3Guard-Gen-4B` | – | – | $0.05 in · $0.75 out per 1M tokens | 4.4B |
| [gliguard LLMGuardrails 300M](/models/fastino-gliguard-llmguardrails-300m.md) `fastino/gliguard-LLMGuardrails-300M` | 0.576 | – | $0.0929 per 1M tokens | 208M |
| [granite guardian 3.0 2b](/models/ibm-granite-granite-guardian-3-0-2b.md) `ibm-granite/granite-guardian-3.0-2b` | 0.247 | – | $0.15 in · $0.60 out per 1M tokens | 2.5B |

– = not measured / not timed / not listed.

[All 3 safety models](/models#guardrails)

## Facts

- Size: 4.4B parameters
- License: Apache-2.0
- Reads: Text, up to 32,768 tokens
- Endpoint: generate

## Evaluation

- Accuracy: Run on SIE against the public dataset named in each figure
- Prices: Serverless list prices
- Sources: [Task page runs and their inputs](https://huggingface.co/datasets/superlinked/sie-task-evidence)