For technical details behind these features and a comprehensive overview of our approach to responsible development, please see the Gemma 3 technical report.
Rigorous safety protocols to build Gemma 3 responsibly
We believe that open models require careful risk assessment, and our approach balances innovation and safety by adjusting the intensity of testing to the capabilities of the model. The development of Gemma 3 includes extensive data governance, fine-tuning and alignment with safety policies through robust benchmarking. While thorough testing of more capable models often results in evaluations of less capable models, Gemma 3’s improved STEM performance led to a specific evaluation focused on its potential for abuse in the production of hazardous substances. The results indicate a low level of risk.
As the industry develops more powerful models, it will be important to jointly develop risk-proportionate approaches to safety. We will continue to learn and refine our open model safety practices over time.
Built-in safety for image applications with ShieldGemma 2
Alongside Gemma 3, we are also releasing ShieldGemma 2, a powerful 4B image safety checker built on the Gemma 3 foundation. ShieldGemma 2 provides a ready-made solution for image safety, outputting safety labels across three safety categories: dangerous content, sexually explicit content, and violence. Developers can further customize ShieldGemma to suit safety needs and users. ShieldGemma 2 is open and built to provide flexibility and control, leveraging the performance and efficiency of the Gemma 3 architecture to foster responsible AI development.
Ready to integrate with tools you already use
Gemma 3 and ShieldGemma 2 integrate seamlessly into your existing workflow.
Develop with your favorite tools: Support for Hugging Face Transformers, Ollama, JAX, Keras, PyTorch, Google AI Edge, UnSloth, vLLM, and Gemma.cpp gives you the flexibility to choose the best tools for your project. Start experimenting in seconds: Get instant access to Gemma 3 and start building right away. Explore its full potential in Google AI Studio or download the model through Kaggle or Hugging Face. Customize Gemma 3 for your specific needs: Gemma 3 ships with an improved codebase that includes recipes for efficient fine-tuning and inference. Train and adapt your models using the platform of your choice, including Google Colab, Vertex AI, and even gaming GPUs. Deploy the way you want: 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 applications and infrastructure. Experience optimized performance on NVIDIA GPUs: NVIDIA has directly optimized the Gemma 3 model for maximum performance on any GPU. Sizes from Jetson Nano to the latest Blackwell chips. Gemma 3 is now available in the NVIDIA API Catalog, enabling rapid prototyping with just a few API calls. Accelerate AI development on many hardware platforms. Gemma 3 is also optimized for Google Cloud TPUs and integrates with AMD GPUs via the open source ROCm™ stack. For CPU execution, Gemma.cpp provides a straightforward solution.
“Gemmaverse” of models and tools
Gemmaverse is a vast ecosystem of community-created Gemma models and tools, ready to foster innovation and inspire. For example, AI Singapore’s SEA-LION v3 removes language barriers and facilitates communication across Southeast Asia. INSAIT’s BgGPT is the first large-scale language model for the pioneering Bulgarian language that demonstrates Gemma’s power to support diverse languages. OmniAudio from Nexa AI shows the potential of on-device AI, bringing advanced audio processing capabilities to everyday devices.
To further promote the advancement of academic research, we are launching the Gemma 3 Academic Program. Academic researchers can apply for Google Cloud credits (valued at $10,000 per award) to accelerate their Gemma 3-based research. Application forms open today and will be accepted for four weeks. Please apply from our website.
Try using Gemma 3
Gemma 3 represents the next step in our continued commitment to democratizing access to high-quality AI. Ready to explore Gemma 3? Start here:
Instant exploration:
Try Gemma 3 in full accuracy directly in your browser using Google AI Studio. No setup required. Get an API key directly from Google AI Studio and use Gemma 3 with the Google GenAI SDK.
Customize and build:
Download the Gemma 3 model from Hugging Face, Ollama, or Kaggle. Easily tweak the model to suit your unique requirements using Hugging Face’s Transformers library or your favorite development environment.
Deployment and expansion:
