A little more than a year after the launch of the “Gemma Ai” “open” models designed with the same technology behind Guemini AI, Google Is Family update with GEMMA 3. According to the publication of the blog, these models are intended for use by developers who create Amnesty International applications capable of operating wherever, on anything from the phone to a work station with support for more than 35 languages, as well as the ability to analyze text, images and short videos.
The company claims to be “the best one acceleration model in the world”, outperforming the competition from Llama on Facebook, Deepseek and Openai Performance on a host with a single graphics processing unit, as well as improved possibilities for GPUS in NVIDIA and custom devices. The Gemma 3 Vision Cracks are also upgraded, with high -resolution and non -square images support, while new ShieldGEMA 2 Photo Safety Class Available to use to liquidate both the entry of the image and the output of the classified content as sexual, dangerous or violent.
Last year, it was not clear how much attention would be there in a model like GEMMA, however, Deepseek and others show that there is an interest in artificial intelligence technology with low devices requirements. Despite its demands for advanced capabilities, Google also says, “The improved STEM performance for GEMMA 3 has paid specific reviews that focus on its capabilities to abuse in creating harmful materials; its results indicate a low level of risk.”
What is exactly the “open” or “open source” model remains the subject of discussion, and with GEMMA from Google, focused on the company’s license that restricts what is allowed for people to use, and which has not changed with this new version. Google continues to promote GEMMA with Google Cloud credits, and the Gemma 3 academic program will allow academic researchers to apply for $ 10,000 credit to accelerate their research.