# Gemma
Gemma is [[Google]] DeepMind's family of open-weight [[Large Language Models (LLMs)|LLMs]]. Built on the same research and technology as [[Gemini]], Gemma models are lightweight, open-source, and designed to run locally on consumer hardware. They are available for free use, fine-tuning, and deployment.
## Versions
| Version | Sizes | Notes |
|---------|-------|-------|
| Gemma 1 | 2B, 7B | Initial release (Feb 2024) |
| Gemma 2 | 2B, 9B, 27B | Improved efficiency and quality (Jun 2024) |
| Gemma 3 | 1B, 4B, 12B, 27B | Multimodal (text + image), long context (128K), multilingual (Mar 2025) |
| [[Gemma 4]] | E2B, E4B, 26B MoE, 31B | Reasoning, audio, MoE, 256K context, agentic/tool use, Apache 2.0 (Apr 2026) |
## Key features
- Open weights — downloadable and runnable locally
- Multimodal from Gemma 3 (vision + language); audio added in [[Gemma 4]]
- Long context window (128K in Gemma 3; up to 256K in [[Gemma 4]])
- Strong benchmark performance relative to size
- Available via Ollama, Hugging Face, Google AI Studio, Vertex AI
- Instruction-tuned variants (`-it` suffix) for chat/agent use
- Native function calling and reasoning mode from [[Gemma 4]]
- [[AI Mixture of Experts (MoE)]] variant introduced in [[Gemma 4]]
## Run locally
Via [[Ollama]]:
```sh
ollama run gemma4
ollama run gemma4:e4b
ollama run gemma4:27b
```
## Access
- Google AI Studio (free)
- Vertex AI
- Hugging Face: `google/gemma-4-27b-it`
- Ollama: `gemma4`
## References
- Official Website: https://gemma.ai/
- Google blog (Gemma 3): https://blog.google/technology/developers/google-gemma-3/
- Google blog (Gemma 4): https://blog.google/innovation-and-ai/technology/developers-tools/gemma-4/
- Hugging Face: https://huggingface.co/google
## Related
- [[Gemma 4]]
- [[Gemini]]
- [[Google]]
- [[Large Language Models (LLMs)]]
- [[AI Mixture of Experts (MoE)]]
- [[AI Multimodal]]
- [[Ollama]]