To dive deeper into technical details behind these opportunities, as well as a comprehensive overview of our approach to responsible development, can be found in the article Gemma Technical Report 3.
Rigorous safety protocols enable responsible building of Gemma 3
We believe that open models require thorough risk assessment, and our approach balances innovation with security – tailoring the intensity of testing to the capabilities of the model. The development of Gemma 3 included extensive data management, alignment with our security policies through tuning, and robust benchmarking. Although thorough testing of more efficient models often allows for the evaluation of models with lower capabilities, Gemma 3's improved performance in STEM has prompted detailed evaluations focusing on its potential for misuse in the creation of harmful substances; their results indicate a low level of risk.
As the industry develops increasingly effective models, it will be critical to collectively develop a risk-proportioned approach to safety. Over time, we will continue to learn and refine our security practices for open models.
Built-in security for graphical applications with ShieldGemma 2
In addition to Gemma 3, we are also releasing a game ShieldGemma 2Powerful 4B image security checker built on the foundation of Gemma 3. ShieldGemma 2 provides a turnkey image security solution that displays security labels in three security categories: dangerous content, sexually explicit content, and violence. Developers can further customize ShieldGemma to meet their security and user needs. ShieldGemma 2 is open and built to provide flexibility and control, leveraging the performance and efficiency of the Gemma 3 architecture to promote responsible AI development.
Ready to integrate with the tools you already use
Gemma 3 and ShieldGemma 2 integrate seamlessly with existing workflows:
- Grow with your favorite tools: With support from Hugging Face Transformers, Ollama, JAXKeras, PyTorch, Google's Artificial Intelligence Edge, LazinessvLLM i Gemma.cppyou have the freedom to choose the best tools for your project.
- Start experimenting in seconds: Get instant access to Gemma 3 and start building right away. Discover its full potential in Google Artificial Intelligence Studioor download models via Kaggle Or Face Hugging.
- Tailor Gemma 3 to your specific needs: Gemma 3 ships from revamped code base which provides recipes for efficient tuning and inference. Train and fine-tune your model using your preferred platform, e.g Google Co, Apex AI or even a gaming GPU.
- Deploy your way: Gemma 3 offers multiple deployment options including Vertex AI, Cloud Run, Google GenAI API, Iocal environments and other platforms, giving you the flexibility to choose the best fit for your application and infrastructure.
- Experience optimized performance on NVIDIA GPUs: NVIDIA has directly optimized Gemma 3 models to deliver maximum performance on GPUs of all sizes, from Jetson Nano to the latest Blackwell chips. Gemma 3 is now featured on NVIDIA API Directoryenabling rapid prototyping with just an API call.
- Accelerate AI development across multiple hardware platforms: Gemma 3 is also optimized for Google Cloud TPUs and integrates with AMD GPUs via the open-source ROCm™ stack. To make the processor Gemma.cpp offers a direct solution.
“Gemmaverse” of models and tools
Gemmaverse is a vast ecosystem of community-built Gemma models and tools ready to support and inspire your innovations. For example, AI Singapore SEA-LION v3 breaks down language barriers and promotes communication in Southeast Asia; BgGPT by INSAIT is a pioneering large Bulgarian language model that demonstrates Gemma's cross-language strength; AND OmniAudio Nexa AI showcases the potential of on-device AI, bringing advanced audio processing capabilities to everyday devices.
To further promote breakthroughs in academic research, we are launching the Gemma 3 Academic Program. Academic researchers can apply for Google Cloud funds (worth $10,000 per award) to accelerate their Gemma 3-based research. The application form opens today and will be open for four weeks. Apply our website.
Start with Gemma 3
As part of our ongoing commitment to democratize access to high-quality AI, Gemma 3 is the next step. Ready to discover Gemma 3? Here's where to start:
Instant Exploration:
- Try Gemma 3 with full precision right in your browser – no configuration required – with Google AI Studio.
- Get your API key directly from Google AI Studio and use Gemma 3 with Google GenAI SDK.
Customize and Build:
- Download Gemma 3 models from Face Hugging, To beOr Kaggle.
- Easily customize and adapt the model to your unique requirements using Hugging Face's Transformers library or your preferred development environment.
Deploy and scale:
