DeepSeek V3's recent minor upgrade, achieved through post-training optimization, surpasses GPT-4.5 in mathematics and programming capabilities. Notably, this update also signifies the integration of V3 and R1 into a unified model, with all model weights now fully open-sourced.
