A quantized variant of Google's Gemma 4 26B model, adapted by bg-digitalservices using NVFP4A16 precision — a format designed to reduce memory footprint while preserving model behavior. It handles text-in, text-out tasks and trades some numerical precision for deployment efficiency. The trade-off is typical of quantized models: smaller resource requirements at the potential cost of subtle quality degradation compared to the full-precision original.