From the source
DeepSeek announces DeepSeek-V3.2-Exp, an experimental model built on V3.1-Terminus that introduces DeepSeek Sparse Attention (DSA) for faster and more efficient training and inference on long contexts.
The model is now live on App, Web, and API.
API prices are reduced by more than 50%.
Benchmarks show V3.2-Exp performs on par with V3.1-Terminus.
V3.1-Terminus remains available via a temporary API until October 15, 2025.
Key GPU kernels are released as open source in TileLang and CUDA.
