Google launches Gemma 3: its most advanced open-source AI for a single GPU

Google has officially announced the launch of Gemma 3, the new version of its open-source artificial intelligence model, designed to run efficiently on a single graphics card (GPU)With this development, the company seeks to offer more accessible and flexible AI for developers and companies that require optimized solutions without the need for expensive infrastructure.

The Gemma family has undergone constant evolution. Following the introduction of the first versions in February of last year, which featured models with 2.000 billion and 7.000 billion parameters, and the subsequent arrival of Gemma 2 in May with 27.000 billion parameters, Google is now expanding the series with Gemma 3, a model designed to balance power and efficiency in different environments..

Main features of Gemma 3

Gemma 3 has been designed in four sizes: 1, 4, 12 and 27 million parameters, allowing it to adapt to different hardware and performance needs. Its efficient execution allows it to work on devices with a single GPU, covering everything from mobile to advanced workstations.

One of its most notable features is its support for 140 languages, with out-of-the-box support for over 35 languages, making it easy to create multilingual applications without any additional configuration.

Optimized performance and comparison with other models

According to Google, Gemma 3 has outperformed competing models in performance tests such as Llama-405B from Meta, DeepSeek-V3 and o3-mini from OpenAI, demonstrating its efficiency in executing complex tasks with a single GPU.

The model has been optimized to work especially with nvidia graphics cards, but it can also work with TPUs and other hardware platforms, ensuring their integration into multiple development environments.

Advanced Features and Applications

Among the most important improvements of Gemma 3, its expanded context window of 128K tokens, allowing the processing and analysis of large volumes of information. It also has the capacity to analyze text, images, and short videos, making it a key tool for generative artificial intelligence.

In addition, it incorporates functions such as function calls and structured output, which allows you to automate tasks and develop intelligent assistants within various applications.

ShieldGemma 2: An AI Security Solution

Along with Gemma 3, Google has also launched ShieldGemma 2, an AI-based security verifier designed to analyze and label images according to three categories: Dangerous content, sexually explicit material and violenceThis tool allows developers to tailor their security systems to the needs of each application.

Availability and integration on different platforms

Gemma 3 is now available and can be integrated into multiple development environments. Access has been enabled in Hugging Face, Kaggle, Google AI Edge and Vertex AI, in addition to having support in Google AI Studio to facilitate its implementation.

This release reinforces Google's commitment to developing open and accessible AI models. With Gemma 3, the company aims to provide researchers and developers with a powerful, optimized, and versatile tool for a wide range of AI applications.