MODELS
Gemma 3
by Google DeepMind
Overview
Open model family from Google DeepMind for text, image understanding, reasoning, multilingual use, and developer deployment.
Details
Gemma 3 is an open-weight model family from Google DeepMind. Official Google sources describe it as a lightweight open-model collection based on Gemini 2.0 research, with 1B, 4B, 12B, and 27B variants, 128K context, multilingual support across 140+ languages, multimodal text/image input with text output, visual reasoning, function calling, quantized versions, AI Studio/API-key access, and deployment options including Google Cloud Model Garden and local/runtime tooling such as gemma.cpp.
When to Use
Use Gemma 3 when you need an open-weight Google model family for text generation question answering summarization reasoning or multilingual applications. Use Gemma 3 for image-understanding or visual-reasoning workflows where text/image input and text output are sufficient. Use Gemma 3 when evaluating deployable model sizes from small 1B variants through larger 27B variants including local/runtime or Google Cloud deployment paths.
Getting Started
- Review the Gemma overview at https://ai.google.dev/gemma and the official docs at https://ai.google.dev/gemma/docs.
- Read the Gemma 3 model card for supported modalities
- context length
- intended uses
- evaluations
- limitations
- and safety considerations.
- Choose a variant such as 1B
- 4B
- 12B
- or 27B based on your deployment and task requirements.
- Use Google AI Studio/API-key access
- Google Cloud Model Garden guidance
- or the official gemma.cpp repository depending on your runtime target.
- Check the Google Cloud pricing page if you plan supervised fine-tuning of Gemma 3 IT variants.
Key Features
- •Open-weight Gemma 3 model family authored by Google DeepMind.
- •Variants documented in Google Cloud sources include 1B
- •4B
- •12B
- •and 27B models.
- •Supports multimodal text/image input and text output according to the official model card.
- •Google launch material states 128K context
- •140+ language support
- •visual reasoning
- •function calling
- •and quantized versions.
- •Deployment and usage paths are documented through Google AI for Developers
- •Google Cloud Model Garden
- •and the google/gemma.cpp repository.
Capabilities
- •text generation
- •question answering
- •summarization
- •reasoning
- •multilingual text
- •image understanding
- •visual reasoning
- •function calling
- •local deployment
- •cloud deployment
- •supervised fine-tuning
Last updated Jul 15, 2026