DeepSeek V4.1 Flash Launches: 552B MoE Model Outperforms V4 Pro

Chinese AI company DeepSeek has officially released the DeepSeek V4.1 Flash model, the smallest version in its new model architecture series, featuring native multimodal visual understanding. It is a 552B parameter MoE model with a new Causal-Encoder-Decoder structure, asymmetric input and output, activating only 8B for input and 16B for output, making costs significantly lower than similar-sized models.

In benchmark tests, V4.1 Flash outperformed flagship models including DeepSeek V4 Pro. The new model greatly reduces KV Cache size, cutting HBM demand to one-fourth and SSD demand to one-eighth compared to the previous generation, significantly lowering cache-hit costs.

V4.1 Flash is now available on DeepSeek API with native multimodal support. Users can call it by changing the model name to deepseek-flash. Tencent WorkBuddy, CodeBuddy, and OpenCode have fully integrated the model. DeepSeek has also lowered pricing with peak and off-peak rates, where off-peak pricing is half of peak pricing.