Google releases Gemma 4 QAT checkpoints to shrink on-device AI models for phones, laptops, and consumer GPUs

Google DeepMind released new Gemma 4 checkpoints optimized with Quantization-Aware Training to reduce memory needs while preserving model quality. The update includes support for Q4_0 and a mobile-focused quantization format that can shrink Gemma 4 E2B to a 1GB memory footprint.

The release reflects a broader push to make capable AI models run locally on everyday devices rather than relying only on cloud inference. Better compression with limited quality loss could accelerate edge AI adoption across mobile apps, laptops, and privacy-sensitive use cases.

Gemma 4 QAT models: Optimizing compression for mobile and laptop efficiency