- Gemma 3 is Google's new open AI model, designed to run efficiently on a single GPU.
- Available in 1, 4, 12 and 27 million parameters, it allows adaptability according to the user's needs.
- It offers support for 140 languages and advanced features such as image, text, and video analysis.
- Includes ShieldGemma 2, a security tool that classifies images based on their content.
Google has unveiled its new series of open AI models, Gemma 3. This version offers significant performance optimizations, allowing it to run on a single graphics processing unit (GPU) . With this strategy, the company aims to facilitate access to advanced AI models without requiring expensive infrastructure.
The Gemma family was launched in February of last year with models handling 2 and 2.000 billion parameters , and in May it was updated with Gemma 7.000, reaching 27.000 billion parameters . Now, with Gemma 3, Google is expanding its offering with lighter, more versatile, and more efficient options for developers.
New capabilities and improvements in AI
Gemma 3 has been designed to ensure optimal performance across a range of devices, from computers to smartphones. It is offered in variants with 1, 4, 12, and 27 million parameters , allowing its implementation to be tailored to different requirements.
One of the most notable aspects of this model is its foundation in the same technology behind the Gemini 2.0 , ensuring superior performance. According to LMArena ranking tests, Gemma 3 outperforms models like Meta's Llama-405B, DeepSeek-V3, and OpenAI's o3-mini , especially in Nvidia GPU -based environments.
Furthermore, this new model is optimized to support 140 languages , with immediate availability in more than 35 languages , facilitating its integration into different markets.
Advanced features and automation
Among its main features, Gemma 3 allows the creation and analysis of content in multiple formats , including text, images, and short videos. Its 128K token context window provides greater information processing capacity.
It also has the ability to execute function calls and structured output , which facilitates task automation and the implementation of interactive experiences within applications.
ShieldGemma 2: Image Security Verification
Along with the launch of Gemma 3, Google has introduced ShieldGemma 2 , an AI-powered security checker. This 4B parameter model analyzes images and categorizes them into three content categories: dangerous, sexually explicit, and violent . This allows developers to customize their security systems to meet the specific needs of their users.
This tool seeks to ensure responsible and safe use of AI, promoting best practices in the implementation of artificial intelligence models in various applications.
Accessibility and integration
Gemma 3 is now available for integration with multiple platforms and development environments, including Hugging Face, Kaggle, Google AI Edge, and Vertex AI . It can also be used from Google AI Studio to facilitate its deployment in various AI projects and experiments.
As part of its commitment to the responsible development of AI, Google has conducted a rigorous risk assessment , ensuring that its models comply with security protocols and standards aligned with technological responsibility.
This new release represents an important step towards the democratization of open artificial intelligence, providing high-performance tools accessible to any user or developer without the need for complex infrastructures.