model · Google DeepMind

Gemma 4

Gemma 4 is Google's latest family of open models, spanning tiny edge variants (a few GB, runnable on a phone) up to a 31B dense model and a 26B mixture-of-experts. It is multimodal (text plus images), handles 140+ languages, has a 256K context window, and ships in both GGUF and Apple's MLX formats. Open licensing makes it free to run and ship. Pick Gemma when you want Google-grade quality locally and a size that matches your machine.

  • free
  • self-host
  • Commercial

What it can help with

  • multimodal
  • edge deployment
  • local inference
  • large language model
  • image input
  • high context window
  • open licensing
  • phone-friendly