Gemma 3, Google’s latest open AI model, offers cutting-edge capabilities while being optimized for efficient performance on a single GPU or TPU. Building on the success of its predecessors, Gemma 3 delivers superior language support, advanced text and visual reasoning, and an expanded 128K-token context window.
Available in multiple sizes (1B, 4B, 12B, and 27B), it allows developers to choose the best fit for their needs.
Gemma 3 outperforms competing models like Llama-405B, DeepSeek-V3, and o3-mini in human preference evaluations. It supports over 140 languages, making it ideal for global applications.
With official quantized versions, it reduces computational requirements while maintaining high accuracy. Its function-calling capability enhances automation, enabling AI-driven workflows.
Google has implemented rigorous safety measures, ensuring Gemma 3 meets high ethical standards. ShieldGemma 2, a new 4B image safety checker, accompanies the launch to help developers detect harmful content.
Gemma 3 seamlessly integrates with popular tools like Hugging Face, Ollama, PyTorch, and Google AI Edge, making AI development more accessible. Optimized for NVIDIA GPUs, Google Cloud TPUs, and AMD ROCm, it provides flexibility for diverse hardware environments.
To support academic research, Google is launching the Gemma 3 Academic Program, offering $10,000 in Google Cloud credits to selected researchers. Developers can explore Gemma 3 instantly through Google AI Studio, Hugging Face, or Kaggle, fine-tune it for custom needs, and deploy it via Vertex AI, Cloud Run, or NVIDIA’s API Catalog.
Gemma 3 marks a significant step in democratizing AI, empowering developers to create innovative, responsible AI applications efficiently.
