MiniMax founder details strategy: prioritizing efficiency and open source path to AGI

MiniMax founder Yan Junjie recently discussed the state of large model development with Lawrence Burns of Scottish Mortgage, noting that China's leading models have achieved about 95% of top U.S. model performance while costing only one-tenth the expense. Despite this, Yan Junjie acknowledged that closing the final performance gap remains a significant challenge, asserting that both regions are still in the early stages toward Artificial General Intelligence (AGI). To manage compute constraints, MiniMax centers its strategy on extreme efficiency gains, reinvesting savings from optimized infrastructure into post-training. The company, which has an annualized revenue run rate exceeding $800 million, with enterprise business accounting for over 80% of that, plans to drive revenue growth beyond cost increases through technological innovation. MiniMax's product portfolio includes the M series for coding and agents, and the H series for video generation, while actively developing the M3Pro model. Yan Junjie stated that the path to AGI requires AI to generate entirely new scientific theories. Regarding ecosystem development, he posits that the U.S. favors capital-intensive closed systems, whereas China's approach emphasizes high efficiency for collaborative systems. Consequently, MiniMax remains committed to open sourcing its technology until AGI is realized, arguing this prevents monopolization and satisfies enterprise needs for data privacy.