Gemma 4 Developer Hub
Everything you need to run, deploy, and debug Gemma 4 models — from edge devices to production clusters.
Gemma 4 Developer Hub
Welcome to gemma4.dev — the independent community reference for Google's Gemma 4 model family.
What's here
Getting Started
Choose your runtime and run your first Gemma 4 model in under 5 minutes.
Model Reference
Compare E2B, E4B, 26B A4B, and 31B across context, VRAM, and use-case fit.
Concepts
Thinking mode, multimodal inputs, quantization, and prompt formatting.
Quick paths
| I want to… | Start here |
|---|---|
| Run locally on my laptop | Ollama quickstart |
| Deploy to production | vLLM guide |
| Use the smallest model | E2B specs |
| Enable reasoning / thinking | Thinking mode |
| Fix a specific error | Error hub |
Model family at a glance
| Model | Parameters | Context | Min VRAM |
|---|---|---|---|
| Gemma 4 E2B | 2B (dense) | 8K | 1.4 GB |
| Gemma 4 E4B | 4B (dense) | 32K | 3.2 GB |
| Gemma 4 26B A4B | 26B (MoE, 4B active) | 128K | 16.4 GB |
| Gemma 4 31B | 31B (dense) | 256K | 24 GB |
About gemma4.dev
gemma4.dev is an independent community project by Midnight Lab. It is not affiliated with Google. All guides are tested against real hardware.