Google’s new open model based on Gemini 2.0

For the deepest diving in the technical details behind these capabilities, in addition to a comprehensive overview of our responsible development, see the technical report.

Strict security protocols for the construction of GEMMA 3 with responsibility

We believe that open models require an accurate evaluation of the risks, and our innovation approach balances with safety – the intensity of the test to sew the capabilities of the model. The development of GEMMA 3 included large -scale data governance, align with our safety policies through accurate evaluation and strong standard measurement. While the comprehensive test of the most capable models to report our evaluation of the lowest ability to do so, the STEM’s support for Gemma 3 has led to specific reviews that focus on its potential for misuse in creating harmful substances; Its results indicate a low level of risk.

With the development of the industry more powerful models, it will be important to develop a collective way of risk suitable for safety. We will continue to learn and refine our safety practices for open models over time.

Coamed safety of photo applications with shieldgemma 2

Besides GEMMA 3, we also launch SHIELDGEMMA 2, which is a strong 4B Safety Auditor based on GEMMA 3. SHIELDGEMA 2 provides a ready -to -safety solution, and take out safety stickers across three security categories: dangerous content, sexually clear and violent. Developers can allocate ShieldGEMA to meet safety needs and users. ShieldGEMA 2 is open and designed to give flexibility and control, and to take advantage of the performance and efficiency of the GEMMA 3 to enhance the development of responsible artificial intelligence.

Ready to integrate with the tools you already use

Gemma 3 and ShieldGEMA 2 are smoothly integrated into the current workflow:

• Development with your favorite tools: With the support of embrace facial transformers, ollama, Jax, Keras, PyTorch, Google Ai Edge, Unloth, VLLM and Gemma.CPP, you have flexibility in choosing the best tools for your project.
• Start experimentation in seconds: Get immediate access to GEMMA 3 and start building immediately. Explore its full potential in Google Ai Studio, or download the models through Kagge or Huging Face.
• Gemma 3 allocate to your specific needs: Gemma 3 is charged with a renovated code base that includes recipes for effective control and reasoning. Train and adapt the model with your favorite basic system, such as Google Colab, Vertex AI, or even your games processing unit.
• Spread your way: GEMMA 3 offers multiple post options, including Vertex Ai, Cloud Run, Google Genai API, ICAL environments and other platforms, allowing you to flexibility to choose the best convenience for your application and infrastructure.
• Experience optimal performance on Nvidia GPUS: NVIDIA has improved gemma 3 models to ensure that you get the maximum performance of graphics processing units of any size, from Jetson Nano to the latest Blackweell chips. GEMMA 3 is now shown on NVIDIA API, which allows rapid initial models with API only.
• Hurry up the development of artificial intelligence across many hardware platformsGemma 3 has also been improved for Google Cloud TPU and integrates with AMD graphics processing units via Rocm ™ open source. To implement the CPU, GEMMA.CPP provides a direct solution.

“GEMMAVERSE” of models and tools

Gemaeverse is a wide ecosystem for GEMMA models and tools created by society, and it is ready to operate and inspire your innovation. For example, SEA-Lion V3 of Ai Singapore broke linguistic barriers and enhances communication throughout Southeast Asia; The Nexa Ai is offered to offer AI’s capabilities to devices, bringing the advanced sound processing possibilities for daily devices.

To increase the enhancement of academic research breakthroughs, we launch the GMMA 3 academic program. Academic researchers can apply for Google’s cloud credit ($ 10,000 per award) to accelerate their GEMMA 3 research. It opens the application form today, and will remain open for four weeks. Submit an application on our website.

Start with GEMMA 3

As part of our continuous commitment to the democratic character to reach high -quality artificial intelligence, GEMMA 3 represents the next step. Are you ready to explore GEMMA 3? Here’s the starting place:

Immediate exploration:

• Try GEMMA 3 full accuracy directly in your browser – no need to set up – with Google Ai Studio.
• Get the API directly from Google Ai Studio and use Gemma 3 with Google Genai SDK.

Allocation and construction:

• Download the Hugging Face 3, OLLAMA or KAGGLE.
• It is easy to adjust the form and adapt it to your unique requirements with the Huging Face Library, or your favorite development environment.

Publishing and size: