How to Run Gemma 4 on Ollama (All Sizes Explained)

Gemma 4 is Google’s latest open-weight model family, released in April 2026. It comes in four sizes — E2B, E4B, E12B, and E27B — all natively multimodal, meaning they handle text and images without any additional setup. The smaller variants (E2B and E4B) run comfortably on a laptop GPU, making Gemma 4 one of the most accessible high-quality models available for local inference.

Gemma 4 Model Sizes

• Gemma 4 E2B — 2.3B parameters. Runs on 4–6GB VRAM. Fast and lightweight, good for simple tasks.
• Gemma 4 E4B — 4B parameters. Runs on 6–8GB VRAM. The sweet spot for most laptop users — strong quality for the size.
• Gemma 4 E12B — 12B parameters. Requires 12–16GB VRAM. Significantly stronger reasoning and coding performance.
• Gemma 4 E27B — 27B parameters. Requires 24GB+ VRAM. Near-frontier performance for a local model.

Start with E4B if you have a modern laptop GPU. Step up to E12B or E27B if you have a desktop GPU with more VRAM.

How to Install Gemma 4 on Ollama

# Pull the default (E4B)
ollama pull gemma4

# Pull a specific size
ollama pull gemma4:e2b
ollama pull gemma4:e4b
ollama pull gemma4:e12b
ollama pull gemma4:e27b

Running Gemma 4

# Interactive chat
ollama run gemma4

# Single prompt
ollama run gemma4 "Explain transformer attention in simple terms"

# Run a specific size
ollama run gemma4:e12b

Using Gemma 4 Vision Features

All Gemma 4 variants support image input natively. Using the Python library:

import ollama

response = ollama.chat(
    model='gemma4',
    messages=[{
        'role': 'user',
        'content': 'Describe what you see in this image',
        'images': ['screenshot.png']
    }]
)
print(response['message']['content'])

This works for diagram analysis, screenshot interpretation, photo descriptions, and more. The E4B model handles vision tasks well for its size.

Gemma 4 for Coding

Gemma 4 shows significant improvements on coding benchmarks compared to Gemma 3. The E12B and E27B variants are competitive with much larger models on standard coding tasks. For coding on a budget GPU, E4B is a solid choice:

ollama run gemma4:e4b "Write a Python function to parse a CSV file and return a list of dictionaries"

What Hardware Do You Need?

Gemma 4 vs Gemma 3 on Ollama

Gemma 4 is substantially better than Gemma 3 at the same parameter count. Key improvements include native multimodal support (Gemma 3 required a separate vision model), significantly better coding performance, and improved instruction following. If you are currently running Gemma 3, upgrading to the equivalent Gemma 4 size is worth doing.

Troubleshooting

• Out of memory: Drop to a smaller size — E2B runs on most hardware
• Slow on CPU: Gemma 4 is designed for GPU inference — CPU-only will be very slow above E4B
• Model not found: Update Ollama to the latest version

Related Guides

• How to Run Llama 4 on Ollama
• How to Use Multimodal Models with Ollama
• Best Ollama Models in 2026