Google releases Gemma 4 QAT models for efficient on-device AI

Google open-sourced quantization-aware training versions of Gemma 4, designed to run efficiently on consumer hardware without sacrificing performance. For indie developers building local-first AI features, this lowers the bar for shipping models on laptops and mobile devices without cloud dependencies.