This is Qwen 3.8 27B with the built-in refusals removed, so it answers questions the standard version would turn down. It runs on your rented server, the chat opens through a protected link, and the API follows the OpenAI format. Removing refusals does not change the law or the rules of use: you answer for your requests and what you do with the output.
In short
Qwen 3.8 Uncensored runs on cards with 24 GB or more. The most affordable suitable card right now is the Tesla V100 32 GB from 23 ₽ per hour, billed by the minute.
Updated 13 October 2026
You need a card with 24 GB of video memory: the weights take about 18 GB, and the context needs room on top. An RTX 4090 with 24 GB or an RTX 3090 will do. An RTX 5090 with 32 GB leaves room for long conversations. If a smaller model is enough, the 14B version runs on 12 GB.
Prices and availability come from the live catalog and update automatically.
Standard models are trained to decline some requests: they answer with a refusal or drift into vague generalities. In this version the refusals were removed with a technique called abliteration, which deletes the part of the weights that drives them. The model's knowledge does not change. Everyday tasks such as code, copy and translation come out much as they do in the standard Qwen 3.8.
Removing refusals does not mean anything goes. The service rules prohibit illegal activity, attacks on other people's systems, spam and phishing, sexual material involving children, extremist material and copyright infringement. The law applies in full, and you answer for what you ask for and publish.
The model runs on your rented server: requests and answers are not sent to outside AI services. If your data is especially sensitive, encrypt the disk yourself.
The API lives at https://<id>-11434.run.gpu.nz/v1, and your rental token goes in as the key. An example request with curl:
curl "https://<id>-11434.run.gpu.nz/v1/chat/completions" \
-H "Authorization: Bearer <token>" \
-H "Content-Type: application/json" \
-d '{"model": "huihui_ai/Qwen3.8-abliterated:27b", "messages": [{"role": "user", "content": "Hi, how do you differ from the standard version?"}]}'In the chat, thinking is off by default, so the model replies straight away. A switch in the chat settings turns it on. The context is 16,384 tokens. If you need another model from Ollama, use Open WebUI + Ollama, and for memory sizing see the VRAM guide for LLMs.