Zhipu and MiniMax stake opposite survival bets: quality pricing vs inference efficiency

AccBalanced · x · 2026-08-31

China's first two listed AI model companies have laid out mirror-image survival theories. Zhipu treats model quality as the asset: its annual report opens with the formula "AGI commercial value = intelligence ceiling × token consumption," backed by an 83% API price increase early this year that demand absorbed.

MiniMax's counter-formula, stated on its August 26 earnings call: most post-training compute is inference, so inference efficiency determines how many experiments a lab can run and thus how fast models improve — cheap inference is a production function for intelligence. Both claims are testable: MiniMax's 3-trillion-parameter M3 Pro must show efficiency yields a competitive frontier model, and Zhipu reports H1 results August 31. The question for anyone allocating across China's AI sector: does quality convert into price, or efficiency into speed?

Original post →

More from Companies & People

Companies & People channel →