Physix Frontier · News Briefing Card (TMTPost · Oct 27, 2026)

MiniMax M3.1 Flash Faces Benchmark Memorization Doubts

KEY FACTS

  • On October 19, MiniMax released its new model M3.1 Flash.
  • The evaluation firm AA gave M3.1 Flash a composite intelligence index score of 42, 13 points higher than M3.
  • Tests by a blogger showed that M3.1 Flash scored near perfect on old questions and near zero on new questions.
  • Within 72 hours of launch, M3.1 Flash processed over 100 trillion tokens.
  • M3.1 Flash's API input and output prices are $0.15 and $0.60 per million tokens, respectively.

KEY DATA

42 pointsComposite Intelligence Index
100 trillionTokens Processed in 72 Hours
$0.15 per million tokensInput Price
$0.60 per million tokensOutput Price

PHYSIX OBSERVATION

MiniMax is rapidly acquiring users through low prices and free strategies, but the memorization controversy exposes the model's weak generalization ability. If evaluation scores rely on memorizing old questions, real-world applications could fail. The industry needs to guard against leaderboard gaming, and users should pay more attention to performance in real scenarios.

Source: TMTPost report

MiniMax M3.1 Flash Faces Benchmark Memorization Doubts | Physix Frontier Briefing