Google recently unveiled its advanced Gemma 4 AI models, designed for on-device use, specifically targeting smartphones. Competing with GLM5 and Qwen3.5, Gemma’s flagship is its closed Gemini model, set to challenge OpenAI and Anthropic. The 2B and 4B parameter versions, weighing only 4.2GB and 5.9GB, optimize for devices with 12GB RAM, forming the backbone of the upcoming Gemini Nano smartphone models. Notable enhancements include improved reasoning, math skills, and rapid responses, with the Fast model boasting up to 4x speed over predecessors and 60% less battery consumption on TPUs. Developer access to Gemini Nano 4 is available through the AICore Developer Preview, enabling early integration into Android apps. Preliminary tests indicate that Nano 4 Full excels at accuracy, while Nano 4 Fast is quicker but more verbose. Overall, Gemma 4 aims to empower on-device AI processing, balancing speed, accuracy, and user privacy, promising significant advancements in the AI landscape.
Source link
Share
Read more