News Summary

DeepSeek has introduced the V3.2-Exp, an experimental AI model featuring a ‘sparse attention’ technique aimed at significantly reducing computational overhead while maintaining performance in long-context scenarios. The new model, which includes the DeepSeek Sparse Attention (DSA) system, is designed to lower API costs by half, reducing the cost to $0.028 per million input tokens. This approach not only demonstrates architectural efficiency but also aims to expand adoption by lowering entry costs for developers and enterprises. The model is already available via DeepSeek’s application, web portal, and API, signaling a shift towards more cost-effective AI operations. DeepSeek’s strategy highlights its commitment to providing efficient and affordable AI solutions in the competitive landscape of AI research.