Just over a year after the release of two “open” GEMMA AI models built of the same technology for her AI Gemini, Google Gemma family update 3. According to the post on the blog, these models are intended for employ by programmers creating AI applications capable of starting wherever they are needed, from the phone to a workstation with support for over 35 languages, as well as the ability to analyze text, images and compact films.
The company claims that this is “the best one -way model in the world”, exceeding the competition from Facebook, Deepseek and OpenAI with a host performance with one GPU processor, as well as optimized GPU starting capabilities and dedicated AI Nvidia equipment and dedicated AI equipment. The Gemma 3 vision code Shieldgemma 2 image safety classifier It is available for filtering both the image input and outputs for content classified as sexually clear, perilous or violent.
Last year, it was not clear how much interest would be in the model like Gemma, but the popularity of Deepseek and others shows that the interest of AI Tech with lower hardware requirements. Despite the claims about advanced capabilities, Google also says: “The improved performance of STEM Gemma 3 prompted specific assessments focused on its potential improper use in creating harmful substances; Their results indicate low risk. “
What exactly is the “open” or “open source” AI remains the subject of the debate, and thanks to Gemma Google, which focused on the license of the company that limits what people can employ it, which has not changed with the novel version. Google is still promoting Gemma from Google Cloud Credits, and the GEMMA 3 academic program will allow academic researchers to apply for loans worth USD 10,000 to accelerate their research.