According to CoinMeta, MiniMax M3.1 Flash has sparked controversy over cheating in tests. The large model evaluation blogger "Cat Boss" noticed that the model scored nearly full marks on old questions but almost zero on new ones. The model consumes tens of thousands of token per question on average, yet some difficult questions could be answered correctly with just about 10,000 token. This abnormal performance raises suspicions that the model may have memorized the answers to some questions in advance. The "Zhi Zhi" large model ranking list, maintained by Cat Boss, uses a private question bank that is not made public and mainly tests logical, mathematical, and programming abilities. In this round, the median score for M3.1 Flash reached 51.76, more than doubling from the previous generation's M3 score of 23.86 and approaching the latest DeepSeek V4.1 Flash score of 56.03. Some developers suspect that the model may have been specifically trained for old questions without truly mastering the problem-solving methods, a practice known as benchmaxxing.