DeepSeek V4.1 Flash Deployed on China's Supercomputing Network with Asymmetric MoE Architecture

DeepSeek V4.1 Flash has launched on China's National Supercomputing Internet platform, offering API access to a 552B-parameter MoE model. Featuring an asymmetric Causal-Encoder-Decoder design, it activates only 8B parameters on input and 16B on output, while cutting KV Cache HBM footprint to one-fourth and SSD demand to one-eighth. The architectural shift substantially curbs inference and caching expenses, easing deployment economics for agent-driven and long-context applications.