Google Releases Gemma 4 Open Models, Enhancing Performance for Mobile and PC

The E2B and E4B models aim to maximize computational and memory efficiency for mobile and IoT devices. These models are said to be capable of executing full offline processing with near-zero latency on edge devices such as smartphones, Raspberry Pi, and Jetson Nano. They also feature audio and vision understanding, supporting real-time edge processing.

The 12B, 26B, and 31B models are positioned to dramatically improve "intelligence per parameter" for PCs. Optimized for consumer GPUs, they are designed to allow students and researchers to utilize workstations as local-first AI servers. Advanced reasoning capabilities for IDEs, coding assistants, and agentic workflows have also been enhanced.

In terms of functionality, native support for function calling enables the construction of autonomous agents capable of planning, app manipulation, and task completion. Additionally, Google aims to provide a multilingual experience that understands cultural context beyond simple translation, as well as performance improvements for specific tasks through fine-tuning within existing frameworks. Regarding security, it is explained that rigorous infrastructure security protocols, similar to those used for proprietary models, have been applied.