Gemma 4 31B Running Fastest on SambaCloud

Gemma 4 31B is Google DeepMind's most capable dense open model to date — and it's running fastest on SambaCloud. Try it today for reasoning, coding, and agentic workflows on SambaCloud.

• Gemma 4 31B is an open-weight (Apache 2.0) frontier dense model built for advanced reasoning, coding, and agentic workflows
• Top benchmarks: 85.2% MMLU Pro, 89.2% AIME 2026 (no tools), 80.0% LiveCodeBench v6, 84.3% GPQA Diamond, Codeforces ELO 2150
• Gemma 4 31B is an open multimodal model and supports text and image input with text output.
• Available as a preview model on SambaCloud playground or API with the model name gemma-4-31B-it

It's the largest dense model in Google's Gemma 4 family, built from the same research foundation as Gemini 3.

Read the official Gemma 4 announcement here.

SambaCloud runs Gemma 4 31B more than 30% faster than the next provider and miles ahead of the rest. The fastest place to run Gemma 4 31B, verified by Artificial Analysis.

Why Use Gemma 4 31B?

Gemma 4 31B brings frontier-class reasoning to an open-weight model that's small enough to fine-tune and deploy on accessible hardware, while SambaCloud delivers it at the lowest latency available. Key strengths include:

Advanced Reasoning

Built as a highly capable reasoner with a configurable thinking mode, Gemma 4 31B scores 89.2% on AIME 2026 (no tools) and 84.3% on GPQA Diamond, with strong multi-step planning and instruction-following. Toggle thinking on or off depending on whether your workload needs deep deliberation or fast turnaround.

State-of-the-Art Coding

Production-grade coding performance with 80.0% on LiveCodeBench v6 and a Codeforces ELO of 2150 — turning a single workstation, or a SambaCloud endpoint, into a frontier-class local-first code assistant.

Native Agentic Capabilities

Native function-calling, structured JSON output, and native system-prompt support let you build autonomous agents that reliably interact with tools and APIs. Pairs naturally with multi-agent frameworks like OpenClaw and CrewAI.

Get Started Quickly with SambaCloud

With just a few lines of Python, you can pass Gemma 4 31B an image and have it reason over what it sees — extracting structured data from a chart, document, or screenshot and returning it as clean JSON. This is the kind of vision-plus-reasoning task that makes the model a strong foundation for document-processing and agentic applications.

import base64
import mimetypes
import os
import urllib.request

from sambanova import SambaNova

# Initialize client using environment variable for security
client = SambaNova(
    api_key=os.environ["SAMBANOVA_API_KEY"],
    base_url="https://api.sambanova.ai/v1",
)

# Fetch a publicly accessible image of a handwritten manuscript and inline it as a data URI.
# (Wikimedia rejects the default urllib User-Agent, so we set a browser-like one.)
source_url = "https://upload.wikimedia.org/wikipedia/commons/c/c3/GeneralRelativityTheoryManuscript.jpg"
req = urllib.request.Request(source_url, headers={"User-Agent": "Mozilla/5.0"})
with urllib.request.urlopen(req) as resp:
    image_bytes = resp.read()

media_type = mimetypes.guess_type(source_url)[0] or "image/jpeg"
image_url = f"data:{media_type};base64," + base64.b64encode(image_bytes).decode("ascii")

system_prompt = "You are a careful transcription assistant built on Gemma 4 31B."

user_prompt = (
    "Transcribe the handwritten text in this image exactly as written, in its "
    "original language, preserving line breaks. Then translate it into English. "
    "Return a JSON object with four keys: 'transcription' (the full original-language "
    "text), 'translation' (the English translation), 'legibility' (high, medium, or "
    "low), and 'notes' (any words you were unsure about, or an empty list). "
    "Respond with ONLY the JSON, no prose."
)

response = client.chat.completions.create(
    model="gemma-4-31B-it",
    messages=[
        {"role": "system", "content": system_prompt},
        {
            # Per Gemma 4's guidance, place the image BEFORE the text for best results
            "role": "user",
            "content": [
                {"type": "image_url", "image_url": {"url": image_url}},
                {"type": "text", "text": user_prompt},
            ],
        },
    ],
    temperature=1.0,
    top_p=0.95,
    top_k=64,
)

print(response.choices[0].message.content)

Check out SambaCloud and explore Gemma 4 31B in the playground, or generate an API key and integrate it into your agentic application today.

Gemma 4 31B brings frontier open intelligence to production. SambaNova delivers it the fastest.