Google releases quantized Gemma 4 models for resource-constrained devices

Google published quantization-aware training (QAT) versions of Gemma 4 designed to run efficiently on mobile phones and laptops without sacrificing much accuracy. For indie developers building AI features on limited hardware, this lowers the barrier to shipping local LLM inference without relying on cloud APIs.