Google Launches Gemma 3: Open-Source AI Models Optimized for Single GPU

The new Gemma 3 models offer multimodal capabilities, support for 140 languages, and enhanced efficiency for on-device applications.

Gemma 3 models come in four sizes (1B, 4B, 12B, and 27B parameters) and are designed to run efficiently on a single GPU or TPU.
The models support text, image, and short video analysis, with a 128,000-token context window for processing large amounts of information.
Pretrained for over 140 languages, the models provide out-of-the-box support for 35 languages, making them highly adaptable for global use.
Google introduced ShieldGemma 2, a 4B parameter image safety checker that filters content for dangerous, explicit, or violent material.
Gemma 3 outperforms competitors like Meta's Llama-405B and OpenAI's o3-mini in human preference evaluations and efficiency benchmarks.