Google launches its latest open AI models

Google Gemma 3 launched the latest version of the AI ​​Open Models family that aims to set a new standard for access to artificial intelligence.

GEMMA 3 is designed on the foundations of the company’s Gemini 2.0 models, to be a weightless, mobile and adaptable to the creation of Amnesty International applications through a wide range of devices.

This version comes hot in the aftermath of the first birthday of Gemma, and it is an impressive adoption of adoption standards. GEMMA models have made more than 100 million downloads and produced the construction of more than 60,000 species of society. It is called “GEMMAVERSE”, and this ecological system refers to a prosperous society aimed at democracy of artificial intelligence.

“The Gemma family of open models is an institution for our commitment to make useful artificial intelligence technology available,” explained Google.

Gemma 3: Features and capabilities

GEMMA 3 models are available in different sizes – 1B, 4B, 12B and 27B parameters – allowing developers to determine a model designed for the requirements of devices and specific performance. These models are faster, even in modest mathematical settings, without compromising jobs or accuracy.

Below are some prominent features of GEMMA 3:

• One series performance: Gemma 3 sets a new standard for one registration models. In assessing the initial human preferences on the LMARNA panel, GMMA 3 outperformed their competitors including Lama-405B, Deepseek-V3 and O3-MINI.

• Multi -language support across 140 languages: After providing food for a variety of fans, GEMMA 3 comes with pre -abilities for more than 140 languages. Developers can create applications related to users with their original tongues and expand the scope of global projects for their projects.

• Sophisticated text and visual analysis: Through the capabilities of advanced texts, image and short thinking capabilities of the video, developers can apply GMMA 3 to formulate interactive and smart applications – with a set of use cases from content analysis to creative workflow.

• Extended context window: By offering a 128k-aken context window, GMMA 3 can analyze and synthesize large data sets, making it ideal for applications that require understanding of expanded content.

• The job function of automating the workflow: With a job calling support, developers can use structured outputs to automate operations and build AIANIC systems.

• Model models for lightweight efficiency: GEMMA offers 3 formal versions, which greatly reduces the size of the model while maintaining the accuracy of the outputs-a reward for developers who improve the portable or resources portable environments.

The advantages of the model performance are clearly clarified in the Chatbot Arena Elo Score. Although the NVIDIA H100 graphics processing unit requests only, the 27B 3B version of GMMA 3 ranks first among the best chat, with ELO 1338 score. Many competitors request up to 32 graphics processing units to provide a similar performance.

One of the strengths of GEMMA 3 lies in its ability to adapt within the current workflow of developers.

• Various tools: Gemma supports 3 libraries and artificial intelligence tools, including embrace facial, Jax, Pytorch, and Google Ai Edge. For improved post, platforms like Vertex AI or Google Colab are ready to help developers start with the least trouble.
• NVIDIA improvements: Whether the use of beginner graphics units such as Jetson Nano or advanced devices such as Blackwell chips, GEMMA 3 guarantees the maximum performance, which increases its simplification through the NVIDIA API catalog.
• Expanding devices support: Beyond NVIDIA, GEMMA 3 with AMD graphics processing units via ROCM staple and supports the CPM with GEMMA.CPP for more diversity.

For immediate experiments, users can access GMMA 3 models via platforms like Hugging Face and Kagge, or take advantage of the Google AI studio to spread the browser.

AI responsible for the official

“We believe that open models require an accurate evaluation of the risks, and our integrity approach balances,” explains Google.

The Gemma 3 team has adopted strict governance policies, by applying a strong refinement and measurement to align the model with moral guidelines. Looking at the models of capacity enhanced in STEM fields, they have undergone specific assessments to mitigate the risk of misuse, such as generating harmful substances.

Google is pushed for collective efforts within the industry to create safety frameworks that suit the growing strong models.

To turn on its role, Google runs the ShieldGEMA 2. 4B photo -safety auditor enhances architecture at GEMMA 3 and comes out of safety stickers across categories such as dangerous content, frank materials and violence. While providing solutions outside the box, developers can customize the tool to meet specially designed safety requirements.

“GEMMAVERSE” is not just a technical environmental system, it is a community -based movement. Projects such as SEA-Lion V3 are from AI Singapore, Ensait’s BGGPT and omniaudio from Nexa AI a certificate of the strength of cooperation within this ecosystem.

To support academic research, Google has also provided the GMMA 3 academic program. Researchers can apply for Google Cloud credits worth $ 10,000 to accelerate their projects that focus on artificial intelligence. Applications are open today and are available for four weeks.

By its access, capabilities and compatibility on a large scale, GEMMA 3 offers a strong argument to become the cornerstone of the artificial intelligence development community.

(Credit image: Google)

See also: Alibaba Qwen Qwq-32B: Endo-reinforcement learning offer

Do you want to learn more about artificial intelligence and large data from industry leaders? Check AI and Big Data Expo, which is held in Amsterdam, California, and London. The comprehensive event was identified with other leading events including the smart automation conference, Blockx, the digital transformation week, and the Cyber ​​Security & Cloud.

Explore the upcoming web events and seminars with which Techforge works here.