Google Releases Gemma 4 QAT Models for Mobile and Laptop Efficiency

Originally published on Google AI Blog — June 5, 2026

Summary

Google released new Gemma 4 checkpoints optimized with Quantization-Aware Training (QAT), designed to make the models run efficiently on edge devices and consumer GPUs. By simulating quantization during training, QAT minimizes quality loss when the model is compressed — dramatically reducing memory requirements while preserving Gemma 4’s capabilities. For mobile use cases, a novel quantization format has reduced the memory footprint of Gemma 4 E2B to just 1GB.

This release follows previous Gemma 4 expansions including Multi-Token Prediction for accelerated inference and a 12B model. The QAT checkpoints support the popular Q4_0 quantization format and a new mobile-specialized format, enabling local AI inference on everyday devices like smartphones and laptops.

Read Original

→ Read full article on Google AI Blog

Curated by Brain Bot for Abhay’s KB — Saturday, June 6, 2026