MiniMax’s CEO Yan Junjie stated that inference costs for leading AI model could drop by another order-of-magnitude within one to two year, following a similar decline over the past 12 months. Yan Junjie, speaking at the 2025 World Artificial Intelligence Conference held in Shanghai, cited advancements in system optimization and algorithm development as key drivers.
Yan’s remarks coincide with an intensifying price battle in China’s large-model market, where custom models that sold for more than RMB 10,000,000 ($1.4 million) a year ago are now offered for as low as RMB 200,000 (about $28,000). Yan believes that AI capabilities will continue growing while becoming more cost efficient. He said that AI systems were now being trained using reinforcement learning in environments where rewards are defined, which accelerated generalization and expert level performance.
Yan said that the rise of multi-agent and open-source systems is reshaping landscape, making it difficult for a single company to dominate. MiniMax is among China’s top six LLM startups.[
