Skip to content

Gemma 3

An open model that runs on a single GPU or TPU

Visit official site

What Gemma 3 is

The model is an open model from Google built on the research and technology behind Gemini 2.0. It is described as lightweight and high-performance, able to run on a single GPU or TPU for developers.

What you can do with it

  • Work with text and visual reasoning in one model
  • Use function calling and structured output for automated tasks
  • Draw on a 128k-token context window for large amounts of information
  • Choose from 1B, 4B, 12B and 27B sizes

Application scenarios

  • Build a real-time translation app on a mobile device
  • Run cross-cultural research with the multilingual capability
  • Add the model to a customer service system for smart replies

Model notes

  • More than 140 languages are supported
  • Quantized versions reduce model size and compute needs
  • Hugging Face, Ollama and Google AI Edge are listed integrations

Pros & cons

✓ What we like

  • Function calling and structured output help automate tasks
  • Integrates with Hugging Face, Ollama and Google AI Edge

! What to watch out for

  • A single GPU or TPU is the described deployment target
  • Fine-tuning and deployment go through the listed external tools

FAQ

What sizes are available?

The material lists 1B, 4B, 12B and 27B.

What is the context window?

The features list a 128k-token context window.

Last reviewed: 2026-11-15

More AI coding tools tools

View all →

How we review