Google Introduces Gemma 3, Open Model To Develop AI

JAKARTA Google introduces Gemma 3, an open base model used to create Artificial Intelligence (AI). This latest version of Gemma is claimed to be lighter and more sophisticated than the previous generation.

Through its official website, Google says that Gemma 3 is made with the same research and technology for the Gemini 2.0 model. That way, its performance will not be much different. However, Gemma 3 is the most sophisticated open model.

To support the maximum AI development, Gemma 3 is able to run quickly on the device. This model works well on mobile phones, on laptops, or on hardware that supports the development of AI applications.

Google states that Gemma 3 can outperform other AI models from competitors, such as Llama-405B, DeepSek-V3, to o3-mini. This belief arises as Gemma 3 is built using the world's best single accelerator model.

"(This advantage is proven) in evaluating initial human preferences on the LMArena ranking board. This helps you create an interesting user experience that can be published on a single GPU or TPU host," Google said, quoted on Wednesday, March 12.

Gemma 3 also has several capabilities that are considered superior to other AI models. This open model supports 140 languages so that it can be used to build applications that can reach globally.

Gemma 3 supports 35 ready-to-use languages. This open model also provides advanced visual reasoning and text capabilities such as analyzing images, text, and short videos. That way, Gemma 3 can be used to create interactive applications.

Gemma 3 is available in four sizes, including 1B, 4B, 12B, and 27B. To maximize Gemma 3 development, Google also provides a sophisticated 4B image security check tool, namely ShieldGemma 2. This tool is a ready-to-use solution for image security.

"(ShildGemma 2) generates security labels in three security categories: malicious content, sexual explicitity, and violence. Developers can adjust ShieldGemma further for their safety needs and users," Google explained.