• by Laurie Sullivan  @lauriesullivan, March 12, 2025

Smaller open-source artificial intelligence (AI) models can run and process data with less energy at lower costs without sacrificing performance or accuracy.

Google's latest AI open-source models, Gemma 3, are based on the same research and technology as Gemini 2.0 that can run faster and cleaner than its prior set of models.

The small footprint can generate advanced text and visual reasoning to analyze images, text, and short videos. It supports more than 35 languages out-of-the-box, with pre-trained support for more than 140 languages. It allows developers and marketers to automate tasks and build agentic experiences, which brands and retailers are expected to create this year to support customization and personalization for consumers.

In early tests, Google said the single-accelerator model outperforms Llama-405B, DeepSeek-V3 and o3-mini in preliminary human preference evaluations on LMArena’s leaderboard.

This helps developers to create engaging user experiences that can fit on one GPU or TPU host, which means it can process at faster speeds with less space.

The Gemma family of open models has more than 100 million downloads within the past year, and 60,000 Gemma variants. The models run fast and directly on devices, from phones to laptops and workstations.

Gemma 3 will integrate with the next version of ShieldGemma. Developers can customize this version, which is designed to detect and flag or block images that contain potentially harmful or explicit content, such as nudity, violence, or offensive depictions. 

The Gemma 3 models come in a range of sizes, allowing developers to choose among 1 billion, 4 billion, 12 billion and 27 billion parameters. 

These ranges permit AI engineers and developers to select the best model for hardware and performance.