MiniMax to Scale M3 Pro Model to 3T Parameters and Launch Domestic Computing Cluster

Chinese AI startup MiniMax announced that its upcoming M3 Pro model will scale to approximately 3 trillion parameters to enhance generalization and intelligence. Additionally, the company is adapting its M3 and H3 models to domestic Chinese chips, with a large-scale domestic computing cluster set to launch soon. A 3-trillion-parameter model represents a massive scale in the LLM landscape, pushing the boundaries of Chinese AI capabilities. The transition to domestic computing clusters reflects China's accelerating efforts to build self-reliant AI infrastructure amid global hardware constraints. The M3 Pro model will expand training in reinforcement learning and long-range tasks. Financially, MiniMax reported a 283.1% year-on-year revenue increase to $117 million for the first half of fiscal year 2026, though it still recorded a net loss of $358 million.