We’re releasing Gemma 4 quantization-aware training checkpoints, reducing memory requirements and improving on-device performance.