At this critical juncture in global AI development, MiniMax founder Yan Junjie recently spoke with Lawrence Burns, an investment manager at Scottish Mortgage, offering an in-depth analysis of the current state of large model development in China and the U.S., the technical paths forward, and the company's future strategic plans.
The Current State of Large Models in China and the U.S.: The Performance Gap and the Challenge of Catching Up from Behind
When discussing the gap between Chinese and U.S. large models, Yan Junjie pointed out that China's leading large models have already reached about 95% of the performance of top U.S. models, while costing only one-tenth as much. However, he also candidly acknowledged that catching up on the final 5% is far harder than closing the previous 95%. Overall, both China and the U.S. are still at an early stage on the path to artificial general intelligence (AGI), and their current pace of progress is roughly comparable.
The Core Strategy for Survival: Driving Post-training Scale Through Extreme Efficiency Gains
Constrained by limited compute resources, MiniMax has always placed efficiency at the core of its strategy. By continuously optimizing model architecture and building its own infrastructure, the company has greatly improved compute utilization efficiency, and it keeps reinvesting the inference costs it saves into post-training. Its core goal is to build a post-training system with leading scale. At the same time, MiniMax's lean team of only about 500 people continues to pursue extremely high organizational and operational efficiency.
In terms of commercialization progress, MiniMax's annualized revenue run rate has now exceeded $800 million, with enterprise business contributing more than 80% of that. Yan Junjie said that in the future the company will continue to reduce costs through technological innovation and drive revenue growth to outpace cost growth, though it has not yet set a specific timeline for profitability.
Founding Motivation and Product Matrix: From Personal Motivation to a Comprehensive Portfolio
Talking about the original motivation for starting the company, Yan Junjie revealed that it came from his maternal grandfather's wish to write a memoir, which the AI technology at the time could not satisfy. After the company was founded, it also established a multimodal technology roadmap from the start.
At present, MiniMax's product line is mainly divided into two directions: the M series, which focuses on coding and agents, and the H series, which focuses primarily on video generation. At the same time, the company is pushing full speed ahead on developing its next-generation model, M3Pro. In Yan Junjie's view, the core criterion for AGI is whether AI can produce entirely new scientific theories and become the main foundation for achieving major scientific breakthroughs.
Different AI Paths for China and the U.S.: Staying Open Source Until AGI Is Achieved
In choosing an ecosystem development path, Yan Junjie believes China and the U.S. will move in different directions: the U.S. ecosystem relies on massive capital to build closed systems, while China's ecosystem relies on high efficiency to build collaborative systems. MiniMax is committed to continuing open source until AGI is achieved. Open source not only attracts a large number of developers to actively participate in model iteration and generate a rich variety of real-world application scenarios, but also perfectly meets government and enterprise customers' core needs for data privacy and security controllability.
He emphasized that AGI technology should not be monopolized by a single entity, and that everyone in society should share in the fruits of intelligent development.
