This is a compact Qwen3 model with 14 billion parameters and refusals removed. The file is about 9 GB, so it fits a 12 GB card such as an RTX 3060, and it answers faster than the larger models, which suits drafting, editing and simple code. For long documents and hard problems, a bigger model is the better choice.
In short
Qwen3 14B Uncensored runs on cards with 12 GB or more. The most affordable suitable card right now is the RTX 3060 12 GB from 6.95 ₽ per hour, billed by the minute.
Updated 13 October 2026
The model needs a card with 12 GB of video memory: the weights take about 9 GB, and the context and buffers take the rest. An RTX 3060 with 12 GB is enough. For more headroom, an RTX 4080 with 16 GB gives you room to spare.
Prices and availability come from the live catalog and update automatically.
Qwen3 14B is a compact model: it answers faster and needs less memory than the 27B version. It is enough for drafting, editing and simple code. For hard code, long reasoning or large texts, pick Qwen 3.8 Uncensored on 24 GB: it has more parameters and its answers are usually more accurate.
The template caps the context at 8,192 tokens, about ten pages of text. That size lets the model and its cache fit in 12 GB. Split long articles and documents into parts. The table compares the 24 GB versions, and the VRAM guide for LLMs explains how memory is counted.
| Model | Weights | Video memory | Context |
|---|---|---|---|
| Qwen3 14B uncensored | about 9 GB | 12 GB and up | 8,192 tokens |
| Qwen 3.8 27B |
| about 18 GB |
| 24 GB and up |
| 16,384 tokens |
| Qwen 3.8 27B uncensored | about 18 GB | 24 GB and up | 16,384 tokens |
The API lives at https://<id>-11434.run.gpu.nz/v1, and your rental token goes in as the key. A Python example with the requests library:
import requests
url = "https://<id>-11434.run.gpu.nz/v1/chat/completions"
headers = {"Authorization": "Bearer <token>"}
body = {
"model": "huihui_ai/qwen3-abliterated:14b",
"messages": [{"role": "user", "content": "Make this email shorter and more polite: [email text]"}],
}
reply = requests.post(url, json=body, headers=headers, timeout=120)
print(reply.json()["choices"][0]["message"]["content"])Answers come without thinking, so the model replies straight away. Repeat requests skip the load time because the model stays in memory while the server runs. To run other Ollama models on the same kind of server, see Open WebUI + Ollama.