"Intelligence vs. Cost": Has MiniMax Cracked the Dilemma?

MiniMax's semi-annual report shows revenue up 283.1% year-over-year, with ARR surpassing $800 million and B-end revenue accounting for 80%. Founder Yan Junjie has proposed a technical roadmap of "minimizing inference costs while maximizing intelligence," arguing that inference efficiency determines the ceiling of intelligence iteration. Through its self-developed MSA architecture, the company has cut the cost per token for long-form text computation to roughly 1/20 of traditional solutions, balancing cost and intelligence while charting a differentiated competitive path for domestic large models.