Google has recently unveiled Gemma 3, a new family of lightweight, open AI models designed to run efficiently on devices like phones, laptops, and workstations. According to an official blog post, Gemma 3 is built on the same technology as Google's Gemini 2.0 models and is available in four sizes (1B, 4B, 12B, and 27B) to suit different hardware and performance needs.

The company claims Gemma 3 outperforms larger models like Llama-405B and DeepSeek-V3 in preliminary evaluations, making it a strong option for developers creating AI applications that require minimal computational resources. The model supports over 140 languages, with out-of-the-box functionality for 35, enabling developers to build multilingual applications. It also features advanced text and visual reasoning capabilities, allowing it to analyse images, text, and short videos.  

Gemma 3 includes a 128,000-token context window, enabling it to process large amounts of information, and supports function calling for automating tasks. Google has also introduced quantised versions of the model, which reduce size and computational requirements while maintaining accuracy.  

Alongside Gemma 3, Google launched ShieldGemma 2, a 4B image safety checker designed to detect harmful content in images across categories like violence and explicit material. Both models integrate with popular developer tools like Hugging Face, PyTorch, and Google AI Studio, allowing for easy customisation and deployment.  

Google emphasised responsible development, stating that Gemma 3 underwent rigorous safety testing, including evaluations for potential misuse in creating harmful substances. The company also announced the Gemma 3 Academic Program, offering $10,000 in Google Cloud credits to researchers using the model for academic projects.  

Gemma 3 is part of Google's broader effort to make AI technology more accessible. Developers can explore the model through Google AI Studio, download it via platforms like Hugging Face and Kaggle, and deploy it using services like Vertex AI or Nvidia's API Catalog.  

Google says that with its focus on efficiency, multilingual support, and accessibility, Gemma 3 aims to empower developers to create AI applications that run directly on devices, reducing reliance on extensive computational resources.