DeepSeek Releases V4.1 Flash With 552B MoE and Asymmetric Compute

DeepSeek has released DeepSeek-V4.1-Flash, a 552-billion-parameter mixture-of-experts model that activates 8B parameters while reading a prompt and 16B while generating. The model introduces a new causal encoder-decoder design and native multimodal visual understanding, positioning a very large model for lower-cost, high-throughput agent workloads.