Qwen3.8-27B (Colombo)

This is Qwen3.8-27B — Alibaba’s compact open model, strong at coding and documents for its size — running on our own GPU server in Colombo. Prompts and answers are processed in Sri Lanka and never leave it: there is no fallback to a server abroad, by design.

We run it with a 131K-token context and low reasoning effort by default. On 250 MMLU-Pro questions through this server, that setting scored 84.0% — higher than the model’s own default, using a third of the tokens. A single request streams at about 73 tokens a second, and it supports tool calling.

For scale: 1,000 requests of about 3,000 tokens in and 1,000 out cost $2.40 — a long prompt and a page of answer each. Every rate is published and stays the same from one request to the next. See every model.

Qwen3.8-27B (Colombo) is billed in rupees, on the same monthly invoice as every other model your team uses — no foreign card, no separate Alibaba account, and one key for all of it.

It runs on our own GPU server in Colombo, and only there: prompts and answers are processed in Sri Lanka and are never routed to a server abroad, not even as a fallback. Why that matters.

Use the OpenAI SDK you already have — point it at Roar AI and name the model. Read the docs.

from openai import OpenAI

client = OpenAI(base_url="https://api.roar-ai.com/v1", api_key="roar_live_…")

reply = client.chat.completions.create(
    model="qwen3.8-27b-lk",
    messages=[{"role": "user", "content": "Summarise this contract in plain English."}],
)