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

Google open-sourced quantization-aware training versions of Gemma 4, letting developers run capable language models on mobile and laptops with lower memory and compute demands. For indie builders working on resource-constrained deployments, this removes friction around either paying for API calls or managing heavy model hosting.