Gemma 4 Is Out: Run a Full Agentic AI Stack on Your Laptop for Free (Apache 2.0, Ollama, Function-Calling)

Google just released Gemma 4 — and it changes the math on self-hosted AI agents.

Here's the quick version: a 31B open-weight model ranked #3 globally among open models, with native function-calling built in, a 256K context window, and an Apache 2.0 license that lets you use it commercially without restrictions. You can run it on your laptop right now using Ollama.

What Makes Gemma 4 Different for Builders

1. Native function-calling — no fine-tuning required

Most open models need fine-tuning or careful prompt engineering to reliably call tools. Gemma 4 has native support for function-calling, structured JSON output, and system instructions baked in.

2. Apache 2.0 — genuinely commercial-use open

Llama 4 has commercial-use restrictions above a certain usage threshold. Gemma 4 ships under Apache 2.0 — no usage caps, no restrictions, no "contact Meta for a license."

3. Runs where your users are

Four model sizes cover the whole spectrum:

• E2B / E4B (edge) — runs on Android, iOS, Raspberry Pi 5. Under 1.5GB RAM for E2B. 128K context.
• 26B MoE — latency-optimized, activates only 3.8B parameters during inference.
• 31B Dense — best quality, runs quantized on a gaming GPU. 256K context.

4. The ecosystem is already there

Day-one support from: Ollama, LM Studio, vLLM, Hugging Face Transformers, llama.cpp, MLX, NVIDIA NIM.

Quick Start: Gemma 4 Agent with Ollama

If you have Ollama installed:

ollama pull gemma4:26b

Basic function-calling agent in Python:

import ollama

tools = [
    {
        "type": "function",
        "function": {
            "name": "search_web",
            "description": "Search the web for current information",
            "parameters": {
                "type": "object",
                "properties": {
                    "query": {"type": "string", "description": "Search query"}
                },
                "required": ["query"]
            }
        }
    }
]

response = ollama.chat(
    model="gemma4:26b",
    messages=[{"role": "user", "content": "What are the latest AI agent frameworks?"}],
    tools=tools
)

if response.message.tool_calls:
    for tool_call in response.message.tool_calls:
        print(f"Calling: {tool_call.function.name}")
        print(f"Args: {tool_call.function.arguments}")

When to Use Gemma 4 vs. Claude Sonnet

Use Gemma 4 when:

• Building offline-first (field tools, low-connectivity, IoT)
• Privacy requirements prevent sending data to external APIs
• Prototyping without API credits
• Building commercial products (Apache 2.0 matters)

Stick with Claude Sonnet 4.6 when:

• You need best reasoning quality on hard problems
• Running multi-agent coordination
• Latency from local inference is a UX problem

The Bigger Picture

Gemma 4 + Ollama is what "AI agent for everyone" actually looks like. Not a cloud subscription. Not a monthly API bill. A capable, commercially-usable agent model that runs on the hardware your users already have.

The builders who figure out the on-device agent use cases in the next 6 months will define a category that nobody has named yet.

AI Builder Club is where we figure out those use cases first.

Join AIBC and share what you build with Gemma 4.