DeepSeek-V3.2-Exp released, training inference efficiency, API synchronous price reduction.

DeepSeek-V3.2-Exp was officially released today, this is an experimental version. As an intermediate step towards the next generation architecture, V3.2-Exp introduces DeepSeek Sparse Attention based on V3.1-Terminus, exploring optimizations and validations for training and inference efficiency of long texts. Currently, the official app, website, and mini-program have been updated to DeepSeek-V3.2-Exp, and the API has also significantly reduced in price.