Search
MiniMax's New Model Faces 'Benchmark Manipulation' Suspicions as Researchers Question Early Access to Evaluation Questions
3 day ago /
Read about 0 minute
Author：小编
On October 19, 2026, Chinese AI company MiniMax unveiled its next-generation model, M3.1 Flash, claiming significant advancements in agent tasks, programming, and cybersecurity. The model's online inference speed nearly tripled compared to its predecessor, while operational costs dropped by approximately 70%. However, controversy arose over its independent evaluation results: researchers suspected that MiniMax may have accessed portions of the test questions from evaluation agency Artificial Analysis during the previous M3 model's assessment and used the content to train M3.1 Flash, resulting in abnormally high scores in specific benchmark tests.
-
C114 Communication Network -
Communication Home
7 X 24 Track global technological trends
