News Summary

Chinese AI company DeepSeek has unveiled its latest AI model, DeepSeek-V3.2-Exp, which incorporates ‘DeepSeek Sparse Attention’ (DSA) to significantly reduce AI processing costs. This new technology aims to slash API prices by 50% and inference costs by up to 60%, maintaining performance while driving innovation in the AI industry. This breakthrough comes as DeepSeek continues to challenge Western AI giants by optimizing computational efficiency, especially given its constraints due to export restrictions. The sparse attention technique, although known for years, is claimed by DeepSeek to achieve fine-grained sparse attention for the first time, potentially reshaping the industry landscape.