On August 26th, during its interim results conference call held in the evening, MINIMAX-W (00100.HK) unveiled its research and development blueprint for the MiniMax M3.1 and MiniMax M3Pro models. Notably, the parameter scale of the M3Pro is projected to surge to approximately 3 trillion (3T), with a heightened emphasis on reinforcement learning and extended task training. This strategic move aims to bolster the model's generalization capabilities and elevate its intelligence quotient. Yan Junjie, the visionary founder and CEO of MiniMax, articulated that the company's forthcoming phase will concentrate on refining model intelligence, enhancing reasoning efficiency, and fortifying infrastructure prowess. Additionally, there will be a concerted effort to expedite the research, development, and launch timelines for both the M-series and H-series models.