Google DeepMind launches Gemma 4 — agentic skills on the edge

AI Impact Summary

Google DeepMind has launched Gemma 4, a family of open models designed for agentic workflows on-device. This release introduces key components like the Google AI Edge Gallery and LiteRT-LM, enabling multi-step planning and autonomous actions across a wide range of hardware, including mobile, desktop, and IoT platforms. The LiteRT-LM library specifically addresses performance concerns with features like minimal memory footprint and constrained decoding, making Gemma 4 suitable for resource-constrained edge devices.