< img src = "/uploads/blogs/12/2f/ib-fqtm7j82i_37c6fc93.jpg" Alt = "Google released the most powerful Shi-model Gemma 3-what is special"/> < P > Google Deepmind command presented Gemma 3 & mdash; A light and productive model based on the same technologies used in the flagship Gemini 2.0. It can work on smartphones, laptops or workstations, allowing developers to create shi app anywhere.
< P > This is referred to in the official google blog.
< H2 > What can a Gemma 3 > 62 ~ 62 < p >Google positions the model as a state -of -the -art and perfect open model. According to the developers, the model is ahead of LLAMA-405B, Deeseek-V3 and O3-mini performance. The main advantage is that the model can be placed on one graphic processor (GPU).
< P > The model allows you to enter the world markets thanks to the full support of 35 languages and previously trained support for more than 140 languages. It allows you to create programs that are able to analyze images, text and short videos, opening up new opportunities for interactive and intellectual applications.
< P > Google paid special attention to a careful risk assessment. GEMMA 3 development included wide data management, coordination with company security policies using accurate adjustment and reliable control estimates.
< P > together with Gemma 3 was also presented Shieldgemma 2, Image Security Program. It derives safety labels by three categories:< Ul > < li > Dangerous content;
< li > frankly sexual content;
< li > Violence.
< P > Developers can self -adjust Shieldgemma according to their security needs and users.
< P > Try Gemma 3 you can in & nbsp; Google Ai Studio. To support academic research, Google also launched a special program where researchers can apply for Google Cloud loans worth $ 10,000 to accelerate their research based on Gemma 3.
~ ~ ~