cortexso/gemma3

Overview

Google developed and released the Gemma 3 series, featuring multiple model sizes with both pre-trained and instruction-tuned variants. These multimodal models handle both text and image inputs while generating text outputs, making them versatile for various applications. Gemma 3 models are built from the same research and technology used to create the Gemini models, offering state-of-the-art capabilities in a lightweight and accessible format.

The Gemma 3 models include four different sizes with open weights, providing excellent performance across tasks like question answering, summarization, and reasoning while maintaining efficiency for deployment in resource-constrained environments such as laptops, desktops, or custom cloud infrastructure.

Variants

Gemma 3

No	Variant	Branch	Cortex CLI command
1	Gemma-3-1B	1b	cortex run gemma3:1b
2	Gemma-3-4B	4b	cortex run gemma3:4b
3	Gemma-3-12B	12b	cortex run gemma3:12b
4	Gemma-3-27B	27b	cortex run gemma3:27b

Each branch contains a default quantized version.

Key Features

• Multimodal capabilities: Handles both text and image inputs
• Large context window: 128K tokens
• Multilingual support: Over 140 languages
• Available in multiple sizes: From 1B to 27B parameters
• Open weights: For both pre-trained and instruction-tuned variants

Use it with Jan (UI)

1. Install Jan using Quickstart
2. Use in Jan model Hub:

   cortexso/gemma3
   

Use it with Cortex (CLI)

1. Install Cortex using Quickstart
2. Run the model with command:

   cortex run gemma3
   

Credits

• Author: Google
• Original License: Gemma License
• Papers: Gemma 3 Technical Report