Google Launches Gemma 3 Family of SLMs

Google today announced its third-generation Gemma 3 family of AI small language models (SLMs) along with the ShieldGamma 2 image safety checker.

“The Gemma family of open models is foundational to our commitment to making useful AI technology accessible,” Google’s Clement Farabet and Tris Warkentin write. “Today, we’re introducing Gemma 3, a collection of lightweight, state-of-the-art open models built from the same research and technology that powers our Gemini 2.0 models. These are our most advanced, portable and responsibly developed open models yet.”

As noted, where Gemini is designed for the cloud, Gemma is designed to run locally against a device’s CPU, GPU, and/or NPU. Gemma 3 is available in a range of sizes–with 1B (billion), 4B, 12B, and 27B parameter versions, each with a 128k token context window–that support smartphones, laptops, and workstations. And Google claims that these are the best AI models to run against at single GPU or NPU.

Gemma 3 supports 35 languages directly, but it’s been pretrained on over 140 languages. These models are multimodal, and they provide advanced text and visual reasoning capabilities and can analyze images, text, and short videos. And the models are agentic, with function calling and structured output support for automating tasks.

Google also announced ShieldGemma 2, which it describes as a powerful 4B image safety checker that’s built on the Gemma 3 foundation.

“ShieldGemma 2 provides a ready-made solution for image safety, outputting safety labels across three safety categories: dangerous content, sexually explicit and violence,” the firm explains. “Developers can further customize ShieldGemma for their safety needs and users. ShieldGemma 2 is open and built to give flexibility and control, leveraging the performance and efficiency of the Gemma 3 architecture to promote responsible AI development.”

You can use Gemma 3 models directly in a web browser with Google AI Studio, and they can be downloaded from Hugging Face, Ollama, and Kaggle. You can learn more on the Gemma 3 website.