Gemma 3n Preview: Powerful, Efficient AI for Mobile Devices

Announcing Gemma 3n Preview: Powerful, Efficient, Mobile-First AI Following the successful launches of Gemma 3 and Gemma 3 QAT—our advanced open-source models designed to run on a single cloud or desktop accelerator—we are excited to take the next step in our mission to make powerful AI more accessible. With Gemma 3, we introduced significant capabilities for developers, and now we are extending this vision to bring real-time AI directly to the devices you use daily, such as phones, tablets, and laptops. To achieve this, we have developed a new, cutting-edge architecture that supports a wide array of applications, enhancing the performance of Gemini Nano. This architecture was crafted in close collaboration with leading mobile hardware companies like Qualcomm Technologies, MediaTek, and Samsung System LSI. It is optimized for fast, multimodal AI, providing users with personalized and private experiences directly on their devices. Gemma 3n is the first open-source model built on this revolutionary shared architecture, offering developers an early preview to experiment with the technology. The same advanced framework will also power the next generation of Gemini Nano, bringing enhanced capabilities to various features in Google apps and our broader on-device ecosystem. Gemini Nano will be available later this year. Gemma 3n is not just another AI model; it stands out for its efficiency and performance. A recent evaluation using the Chatbot Arena Elo scoring system shows that Gemma 3n ranks highly among both popular proprietary and open models, indicating strong user preference. Additionally, the chart below demonstrates its MMLU performance versus model size, highlighting the flexible "mix-n-match" (pretrained) capabilities of Gemma 3n. This chart ranks AI models by Chatbot Arena Elo scores; higher scores (top numbers) indicate greater user preference. Gemma 3n ranks highly amongst both popular proprietary and open models. This chart shows MMLU performance versus model size of Gemma 3n's "mix-n-match" (pretrained) capability. By integrating Gemma 3n into the broader AI landscape, we aim to empower developers to create innovative and impactful applications. Whether you are building sophisticated chatbots, enhancing voice recognition systems, or developing new forms of interactive content, Gemma 3n provides a robust and adaptable platform to meet your needs. Moreover, the mobile-first approach of Gemma 3n ensures that it can run efficiently on a variety of devices, making AI technologies more accessible to a wider audience. This democratization of AI not only improves user experiences but also promotes privacy and data security by keeping sensitive information on-device rather than sending it to centralized servers. As we continue to refine and expand the capabilities of Gemma 3n, we invite the developer community to join us in this journey. Your feedback and creativity will play a crucial role in shaping the future of on-device AI, ensuring that it meets the evolving demands of users and advances the field of machine learning. Stay tuned for more updates and tools to help you get started with Gemma 3n and the next generation of Gemini Nano. Together, we can unlock new possibilities and transform the way AI interacts with everyday technology.