GermanyGoogle has introduced a new on-device AI model called Gemma 4 E2B, specifically optimized for its Tensor Processing Units (TPUs) in Pixel devices. The model, which uses approximately 2.3 billion effective parameters (with embeddings totaling 5.1 billion), allows users to run various AI tasks directly on their smartphones without needing cloud connectivity. This enhances privacy by keeping sensitive data local. Unlike previous models like Gemma 4 12B, this version includes Quantization-Aware Training (QAT) to reduce storage requirements while maintaining performance, making it more accessible to a broader range of devices. The technology enables features such as offline AI agents for travel planning, recipe suggestions, and smart home control, as well as capabilities like transcribing audio and performing complex reasoning across text, images, and audio files. Google plans to extend these capabilities beyond its own apps by providing developers with tools to create locally running applications.
Bias read (Center): The article presents factual information about Google's technological advancements in on-device AI without overtly favoring any political ideology. It focuses on technical specifications, benefits, and future implications of the new AI model, without taking a stance on political issues or policies.
