Qwen 3.8 is Alibaba's latest 27B model: it writes code and prose, analyzes material and handles Russian well. On gpu.nz it starts from one template: rent a server with 24 GB of video memory, wait for the model to load, and open the chat in your browser. The same server exposes an API for your own programs.
In short
Qwen 3.8 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
The model file is about 18 GB, so you need a card with 24 GB of video memory. An RTX 4090 with 24 GB handles the 16,384-token context, and an RTX 3090 does too. An RTX 5090 with 32 GB leaves room for long chats. A 16 GB card will not fit this model.
Prices and availability come from the live catalog and update automatically.
Qwen 3.8 is a general-purpose model. It is handy for code, copy and analysis: write a function, rework a paragraph in a formal tone, compare two documents point by point. It answers in the language of your question, so you can work in Russian or English.
Break a big task into steps: a plan first, then the code, then a check. That usually gives better answers than one large request. The reasons to run a model on your own server are covered on the LLM solution page.
You do not need to install anything: the template starts Ollama, Open WebUI and the Qwen 3.8 weights for you. To set the model up on another server, Ollama needs one command: ollama pull qwen3.8:27b. For a closer look at Ollama and vLLM, read the Ollama and vLLM guide.
The API lives at https://<id>-11434.run.gpu.nz/v1 and follows the OpenAI format. <id> is the part of the Ollama API button's address before the dash. In the API key field, put the token from the «Access token for web interfaces» line on the server page.
from openai import OpenAI
client = OpenAI(
base_url="https://<id>-11434.run.gpu.nz/v1",
api_key="<token>", # the access token from the server page
)
answer = client.chat.completions.create(
model="qwen3.8:27b",
messages=[{"role": "user", "content": "Explain what a Python decorator does."}],
)
print(answer.choices[0].message.content)The template sets a 16,384-token context, roughly twenty pages of English text. That covers a letter, an article or a medium-sized module. Short answers come back in seconds and long ones stream in as they are written; speed depends on the card. The model stays in video memory while the server runs, so you do not wait for it to reload between questions. To work out how much memory a context needs, see the VRAM guide for LLMs.