DeepSeek R1-0528: AIME Accuracy Gain and Single-GPU Model
DeepSeek said R1-0528 raised AIME 2025 math accuracy from 70% to 87.5%, narrowed the gap with o3 and Gemini 2.5 Pro, and added a one-GPU distilled model.
DeepSeek said R1-0528 raised AIME 2025 math accuracy from 70% to 87.5%, narrowed the gap with o3 and Gemini 2.5 Pro, and added a one-GPU distilled model.