MiniMax Bets the Moat Is Cheap Inference

Yan Junjie argues that a compute constraint forced Chinese labs into better engineering, not weaker models.

Speaking on Scottish Mortgage’s Invest in Progress podcast in early November 2026, MiniMax chief executive Yan Junjie said the central problem for Chinese AI companies is driving inference cost low enough that frontier capability reaches ordinary users. He said the constraint “made us better engineers and researchers,” and that closing the final 5% of the gap to US frontier models will be harder than covering the first 95%. He also pledged to keep open-sourcing until AGI is reached. The comparison of matching frontier performance at one-tenth the cost dates from December 2025; he offered no updated figure in this interview.