minimax-m3.1IMAGE
User questions MiniMax M3.1 Flash scores: near-perfect on old problems, near-zero on new ones
A user reports MiniMax M3.1 Flash scores near full marks on old problems but close to zero on new ones, solving old ones in a few thousand tokens while stalling on hard new ones, and questions what such benchmarks measure.
兴雨
@YxzRainyArchived 8 days agoOriginally posted 2026.10.19
兴雨 @YxzRainyOriginally posted 2026.10.19
MiniMax M3.1 Flash 这下有点尴尬了🤣 有人测试，老题接近满分，新题接近 0 分， 难题想半天不会做，老题几千 Token 就搞定了🤡 这跑分到底是在测推理能力，还是测背题能力看😂 如果真是这样，这种基准榜单还有多少参考价值？🤔
x.com/yxzrainy/status/2112208500373999877Archived 2026.10.20 20:15