Google unveils Gemma 3: Open-source AI models that run on a single GPU By Investing.com

Google has released Gemma 3, a groundbreaking collection of open-source AI models that operate efficiently on a single GPU or TPU. Built on the same foundation as Google’s Gemini 2.0, these models range from 1-27 billion parameters, making advanced AI accessible to developers with limited computational resources. Despite modest hardware requirements, Gemma 3 outperforms competitors like DeepSeek, OpenAI, and Meta’s Llama in human preference evaluations. The models support over 140 languages with a 128,000-token context window, enabling sophisticated reasoning across lengthy inputs. Google also introduced ShieldGemma 2, a safety system for image applications that helps identify problematic visual content. Through partnership with NVIDIA, Gemma 3 is optimized for various GPU configurations and integrates with popular AI development platforms including Hugging Face, PyTorch, and others. This release democratizes AI technology by reducing barriers to entry for developers, researchers, and organizations without access to expensive infrastructure.