MiniMax (00100.HK) said on its Aug. 26 interim-results call that R&D on MiniMax
M3.1 and M3Pro will continue; M3Pro’s parameter scale is expected to rise to
about 3 trillion parameters. The company plans to expand reinforcement learning
and long-range task training to improve model generalization and raise the
model’s intelligence ceiling. Founder and CEO Yan Junjie said MiniMax will push
coordinated improvements in model intelligence, inference efficiency and
infrastructure, and accelerate R&D and release cycles for the M and H series.
