What it was trained for

Gemma 3 is Google's family of open, lightweight models built from the same research as Gemini, with multimodal support in larger sizes and designed to run efficiently on a single GPU or modest hardware.

Best for

• ▸On-device and single-GPU deployment
• ▸Local and privacy-sensitive inference
• ▸Fine-tuning for custom tasks
• ▸Multilingual text generation
• ▸Image-and-text understanding in larger variants

Capabilities

Performance & positioning

Strong quality for its size class, delivering competitive results among open lightweight models while remaining efficient enough for local deployment.