DeepSeek Launches V3.2-Exp Model and Slashes API Prices by Over 50%
Chinese AI company DeepSeek has officially launched and open-sourced DeepSeek-V3.2-Exp, an experimental large language model representing a significant step toward its next-generation architecture. The new model introduces DeepSeek Sparse Attention, a fine-grained mechanism designed to enhance efficiency in long-text training and inference without compromising output quality. In benchmarks conducted under aligned training settings against the previous V3.1-Terminus model, V3.2-Exp demonstrated comparable performance across public evaluation datasets. The model is currently available on major platforms such as Hugging Face and ModelScope, with its accompanying technical paper published on GitHub. Alongside this technological release, DeepSeek has updated its applications and developer platforms to support the new model. Furthermore, the company announced a substantial reduction in API pricing, cutting costs by more than 50 percent. This strategic move aims to make advanced AI capabilities more accessible to developers while maintaining high performance standards. The release underscores DeepSeek's continued innovation in optimizing large language models for both efficiency and cost-effectiveness in the competitive AI landscape.
Wire timeline
DeepSeek Launches V3.2-Exp Model and Slashes API Prices by Over 50%
Chinese AI company DeepSeek has officially launched and open-sourced DeepSeek-V3.2-Exp, an experimental large language model representing a significant step toward its next-generation architecture. The new model introduces DeepSeek Sparse Attention, a fine-grained mechanism designed to enhance efficiency in long-text training and inference without compromising output quality. In benchmarks conducted under aligned training settings against the previous V3.1-Terminus model, V3.2-Exp demonstrated comparable performance across public evaluation datasets. The model is currently available on major platforms such as Hugging Face and ModelScope, with its accompanying technical paper published on GitHub. Alongside this technological release, DeepSeek has updated its applications and developer platforms to support the new model. Furthermore, the company announced a substantial reduction in API pricing, cutting costs by more than 50 percent. This strategic move aims to make advanced AI capabilities more accessible to developers while maintaining high performance standards. The release underscores DeepSeek's continued innovation in optimizing large language models for both efficiency and cost-effectiveness in the competitive AI landscape.
TechNode