Archive — CN — technode
DeepSeek Releases V3.2-Exp Experimental Model, Cuts API Prices by Over 50%
DeepSeek has launched the open-sourced DeepSeek-V3.2-Exp experimental model, featuring Sparse Attention for better efficiency in long-text training. It matches performance with the previous V3.1-Terminus model, is available on Hugging Face, and has seen API prices cut by over 50%.
Key points
- DeepSeek released DeepSeek-V3.2-Exp, an experimental large language model.
- Introduced Sparse Attention mechanism for improved efficiency in long-text tasks.
- V3.2-Exp achieved comparable results to V3.1-Terminus in aligned training settings.
- Available on platforms like Hugging Face and ModelScope.
- API prices reduced by over 50% to enhance accessibility.
Why it matters
- The release aligns with ongoing trends in AI focusing on cost-efficiency and performance.
- The trend of open-sourcing AI models has accelerated collaboration in the tech community.
- AI models are increasingly crucial for businesses, reflecting a shift toward AI-driven solutions.
DeepSeek's V3.2-Exp launch signals a significant move towards efficient AI solutions while enhancing accessibility through major API price reductions.