Get Gemma 4
Every official download link, tool, and resource you need — all in one place. Apache 2.0 licensed, 400M+ downloads, 100K+ community fine-tuned variants.
Quickest Way to Start
Install Ollama and run Gemma 4 with a single command
Requires 32GB+ RAM/VRAM for the 31B variant. Use gemma4:e4b for 8GB systems.
Model Downloads
Hugging Face Collection
Official Gemma 4 model weights — all variants (E2B, E4B, 26B, 31B) with quantized and full-precision options.
https://huggingface.co/collections/google/gemma-4-680b4e3622a18fb848e09b8b
Kaggle Models
Gemma 4 on Kaggle — download weights, explore notebooks, and run experiments in hosted environments.
https://www.kaggle.com/models/google/gemma-4
Development Tools
Ollama
Run Gemma 4 locally with a single command. The most popular tool for local inference — supports all variants with automatic quantization.
https://ollama.com
GitHub Repository
Official Gemma C++ implementation. Source code, build instructions, and community contributions for running Gemma 4 natively.
https://github.com/google/gemma.cpp
Which Variant Should You Download?
| Your Hardware | Recommended | Command |
|---|---|---|
| 4-8GB RAM (Mobile/Pi) | E2B | ollama run gemma4:e2b |
| 8-16GB RAM/VRAM | E4B | ollama run gemma4:e4b |
| 24GB VRAM (RTX 4090) | 26B MoE | ollama run gemma4:26b |
| 32-48GB+ VRAM | 31B Dense | ollama run gemma4:31b |
| 80GB (H100) | 31B Full Precision | vllm serve google/gemma-4-31b |
Licensing: Apache 2.0
Gemma 4 is released under the Apache 2.0 license, removing all previous commercial restrictions. You can use, modify, and distribute the models freely for both personal and commercial applications.
This is a significant licensing shift from earlier Google open models, making Gemma 4 the most permissively licensed frontier-class model family available.
Need help choosing? See the deployment guides.
Deployment Guides