Tag: DeepSeek

AI

DeepSeek V3.2-Speciale challenges OpenAI reasoning dominance

DeepSeek R1 achieves 97.3% on MATH-500, outperforming OpenAI o1 while costing only $0.55 per million input tokens. The model utilizes a sparse Mixture-of-Experts architecture to mitigate US chip export restrictions and hardware constraints.