A highly efficient text model optimized with NVIDIA's FP4 quantization, making it lighter and faster to run while preserving the core capabilities of the DeepSeek V4 lineage. It handles a massive 1M token context window, which means it can work across very long documents or conversations without losing track. The trade-off is that quantization can introduce subtle quality degradation compared to full-precision variants.