# Gemma 4 — Full Reference

> Gemma 4 is Google DeepMind's family of open-source AI models, released April 2, 2026, built on Gemini research and distributed under the Apache 2.0 license. Gemma-4.net is an independent, community-maintained guide with hands-on benchmarks and deployment instructions for running Gemma 4 locally. It is not affiliated with, endorsed by, or sponsored by Google or Google DeepMind.

## What Is Gemma 4?

Gemma 4 is Google DeepMind's open-source AI model family, released April 2, 2026 under the Apache 2.0 license. It is built on the same research lineage as Gemini, supports multimodal understanding and a large (256K token) context window, and covers over 140 languages. Because it is Apache 2.0 licensed, it can be downloaded, run locally, fine-tuned, and used commercially without per-token API fees.

Gemma 4 ships in multiple sizes (including the compact E2B and E4B variants and larger models), so it can run on hardware ranging from consumer GPUs and Apple Silicon laptops to multi-GPU workstations, depending on the chosen size and quantization.

Source: https://gemma-4.net/what-is-gemma-4

## Running Gemma 4 Locally

Gemma 4 can be run with several local runtimes:

- **Ollama** — simplest path for most users; one-line model pull and run. https://gemma-4.net/ollama
- **llama.cpp / GGUF** — quantized GGUF files for CPU/GPU inference. https://gemma-4.net/download
- **Apple MLX** — optimized inference on Apple Silicon (M-series). https://gemma-4.net/mlx
- **KoboldCpp** — GGUF-based runtime with a UI. https://gemma-4.net/koboldcpp
- **Unsloth** — fine-tuning and producing quantizations. https://gemma-4.net/unsloth

Hardware requirements scale with model size and quantization level. See https://gemma-4.net/requirements for the VRAM and system requirements per variant.

Source: https://gemma-4.net/deploy

## Downloads (GGUF)

Quantized GGUF builds are available for the compact variants, with file sizes, minimum VRAM, and copy-paste download commands (huggingface-cli / ollama / llama.cpp) on each page. Common quantizations include Q4_K_M (smallest practical, good quality), Q5_K_M, Q8_0, and BF16 (full precision).

- Gemma 4 E2B GGUF: https://gemma-4.net/download/gemma-4-e2b-gguf
- Gemma 4 E4B GGUF: https://gemma-4.net/download/gemma-4-e4b-gguf

(GGUF file sizes are verified against the Unsloth Hugging Face repository.)

## Licensing & Commercial Use

Gemma 4 is distributed under the Apache 2.0 license, which permits commercial use, modification, and redistribution. Always review the official model license terms on the model card before commercial deployment.

## Frequently Asked Questions

- **Is Gemma 4 free to use?** Yes — it is open-source under Apache 2.0 and free to download and run locally.
- **Who made Gemma 4?** Google DeepMind.
- **What hardware do I need?** Depends on the model size and quantization; the compact E2B/E4B GGUF variants run on consumer GPUs and Apple Silicon. See the requirements page.
- **Can I use Gemma 4 commercially?** Yes, under the Apache 2.0 license terms.
- **What is the difference between Gemma 4 and Gemini?** Gemma 4 is the open-source, locally-runnable family; Gemini is Google's proprietary hosted model family. They share research lineage.

Source: https://gemma-4.net/faq

## About This Site

Gemma-4.net is an independent guide maintained by Ethan Lin (Founder & Open-Source LLM Engineer), who benchmarks Gemma 4 quantizations across Ollama, llama.cpp, and MLX. The site is community-maintained and has no affiliation with Google.

Source: https://gemma-4.net/about