Google presents Gemma 3, its new “most capable” open model that can be executed in a single GPU

Google He has presented his new collection of open artificial intelligence models (AI) Gemma 3 that, available in the sizes of 1, 4, 12 and 27 million parameters, it is about more powerful model of the company capable of running in a single graphic unit (GPU), with the aim of helping developers to create AI applications, wherever they need them. ”

Gemma is the family of open source AI models developed by the technological, which was initially presented in February last year in two sizes: 2,000 million parameters (2b) and 7,000 million parameters (7b), and which updated in May in the framework of its annual conference of developers with the launch of Gemma 2, with which It reached a size of 27,000 million parameters (27b) And, with this, a llama performance 3 of the finish line.

Now, Google has expanded the family with its new AI model Gemma 3, which defines as “the most capable model” to execute in a single graphic processing unit (GPU), available in the 1, 4, 12 and 27 million parameter sizes To choose the option that best suits the specific needs of ‘hardware’ and performance.

Thus, these models are designed to run quickly on any devicefrom ‘smartphones’ to laptops or work stations, so they have been devised with the aim of helping developers to Create AI applications “wherever they need them.”

In addition, it is a collection of open and light models that are based on “the same research and technology” that drives its models Gemini 2.0, As Google has detailed in a statement in your blog.

In this sense, the company has stressed that Gemma 3 offers high performance for its sizes, which ensures that It exceeds other models of the sector such as Meta-405b, Deepseek-V3 and O3-Mini de OpenAiin preliminary evaluations in the LMarena classification. Therefore, it allows Create “attractive user experiences” that adapt to a single GPU or TPU host, specifically, with optimized performance in the GPU Nvidia.

Following this line, the new open model of AI is Compatible with 140 languagesand offers immediate support for more than 35 languages, which allows developers to create applications in the language of their users.

On the other hand, Google has pointed out that with Gemma 3 developers can also create in a simple way Applications that analyze short images, text and videos. Also offers a window of 128K context tokens to allow applications to process and understand large amounts of information.

Another of the options that Gemma 3 allows the calls of functions and structured output, with what enables task automationas well as create agent experiences In applications.

Google has also presented ShieldGemma 2its new image safety verifier with 4B size based on Gemma 3. This solution Evaluate the images and generate security labels In three categories, such as dangerous, sexually explicit and violence.

In this sense, developers can use Shieldgemma 2 to integrate it into their applications in order to customize their safety needs for users. Also, take advantage of the performance and architecture of Gemma 3 to promote the responsible development of AI.

In addition to all this, Google stressed that, when developing the model, a “thorough” risk assessment based on your security protocols, including alignment with their security policies through “precise adjustments and solid comparative evaluations.”

With all this, Gemma 3 and Shieldgemma 2 They are already available to integrate into workflows With tools like Hugging Face, Kaggy, Google ai Edge or VerTex AI, among other options. You can also access Gemma 3 capabilities from Google to study.