【Tech 24H】DeepSeek has released the DeepSeek-V3.2-Exp model, an experimental version serving as an intermediate step toward a new-generation architecture. Building upon V3.1-Terminus, V3.2-Exp introduces DeepSeek Sparse Attention, a sparse attention mechanism designed to explore and optimize training and inference efficiency for long-text processing. The model has now been updated across DeepSeek's official app, web platform, and mini-program. Additionally, the API pricing has been significantly reduced, cutting costs for developers by over 50% under the new pricing policy. The DeepSeek-V3.2-Exp model now made its open-source on Huggingface and ModelScope.