DeepSeek-V3.2-Exp Released: Improved Training & Inference Efficiency, API Prices Reduced

The DeepSeek team has released the experimental model DeepSeek-V3.2-Exp. Built upon V3.1-Terminus, it introduces the DeepSeek Sparse Attention (DSA) mechanism, significantly enhancing training and inference efficiency for long-text processing while maintaining performance on par with V3.1-Terminus. DeepSeek-V3.2-Exp is now available on the official app, web platform, and mini-programs. Meanwhile, API prices have been substantially reduced, cutting invocation costs by more than 50%.