Google Unveils Gemma 3, Its Most Advanced Open-Source AI Model Yet

Designed for efficiency and versatility, Gemma 3 can run on a single GPU or TPU and supports over 140 languages while offering advanced multimodal capabilities.

Gemma 3 comes in four sizes, from 1 billion to 27 billion parameters, and is optimized for devices ranging from smartphones to workstations.
The model supports over 140 languages, with pre-trained packages for 35 languages, and excels in text, image, and video analysis.
Gemma 3 introduces a 128,000-token context window, enabling it to process large inputs like full-length books.
The release includes quantized versions for faster performance and reduced computational requirements, making it ideal for mobile and edge devices.
Developers can access Gemma 3 through platforms like Hugging Face, Kaggle, and Google AI Studio, with deployment options ranging from local devices to cloud environments.