Gemma 4 31B

Gemma 4 31B is the biggest model in Google’s open Gemma family, built from the same research as Gemini 3. It is the first Gemma released under the standard Apache 2.0 licence rather than Google’s own terms, and at launch it ranked third among open models on the LMArena text leaderboard. Google’s model card reports 85.2% on MMLU-Pro and 84.3% on GPQA Diamond.

It reads images and video as well as text, with a 256K-token context, and thinks before answering when you switch that on. As a dense 31B model it is small next to the trillion-parameter open models, which shows on long, hard agent tasks — but it is quick and inexpensive for document and image work.

For scale: 1,000 requests of about 3,000 tokens in and 1,000 out cost $1.07 — 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.

Gemma 4 31B is billed in rupees, on the same monthly invoice as every other model your team uses — no foreign card, no separate Google account, and one key for all of it.

Prompts sent to Gemma 4 31B are processed outside Sri Lanka. If a workload has to stay in the country, Qwen3.8-27B (Colombo) runs on our own servers in Colombo.

Google says Gemma 4 was trained on more than 140 languages, but Sinhala and Tamil are not among the 35 it names as supported out of the box. Test it on your own messages first.

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="gemma-4-31b-it",
    messages=[{"role": "user", "content": "Summarise this contract in plain English."}],
)