Google Releases Open-Source AI Model 'Gemma 3'; Capable of Running on a Single GPU

# Google's Open-Source AI Model 'Gemma 3' Unveiled Google introduced 'Gemma 3,' an open-source artificial intelligence (AI) model based on Gemma 2.0, on the 12th (local time). Gemma 3 is a streamlined, open AI model derived from Google's large language model (LLM) 'Gemini.' Gemini serves as the core engine of Google's AI chatbot, Gemini AI. Gemma 3 is capable of running on a single graphics processing unit (GPU), boasting competitive performance against larger models. The model is available in four sizes with parameters of 1 billion, 4 billion, 12 billion, and 27 billion, allowing users to select a model suitable for their device's performance. Clement Farabet, Vice President at Google DeepMind, and Tris Warkentin, Director, described Gemma 3 as "the most advanced portable open model." In performance evaluations, Gemma 3 demonstrated competitiveness. On the open-source AI model performance comparison platform, LMArena, Gemma 3 ranked among the top 10, outperforming large models such as Meta's LLaMA-402B, DeepSeek-V3, Alibaba's Qwen 2.5 Max, and OpenAI's o3-mini. The 27-billion-parameter model achieved a score of 1339 in the LMSys Chatbot Arena Elo rating, earning high marks. Gemma 3 showed notable strengths in creative capabilities, supporting over 140 languages and handling not only text but also images and short videos. Additionally, it can process up to 128,000 tokens at a time, which is more than four times higher than the 32,000 tokens supported by Alibaba's latest lightweight model, QwQ-32B. However, Gemma 3 revealed limitations in document analysis. It failed to complete the analysis of a 47-page International Monetary Fund (IMF) report. The direct copy-input method used in Grok-3 also proved ineffective. In conjunction with Gemma 3, Google also unveiled 'ShieldGemma 2.' ShieldGemma 2 is an AI model designed for image safety review, assessing the appropriateness and analyzing the potential harmfulness of image content.