minimax-m3.1POST
MiniMax M3.1 Flash accused of cheating: still answers the original question after the correct answer was changed
Claims M3.1 Flash jumped from 17% to 56% on AA-Omniscience, that a researcher suspects it trained on undisclosed test items, and that it still output the old answer after a public question's correct answer was altered.
gshota
@gshota666Archived 7 days agoOriginally posted 2026.10.20
gshota @gshota666Originally posted 2026.10.20
🚨 MiniMax 刚发布新模型，就被质疑“作弊”！ M3.1 Flash 在 AA-Omniscience 测试中的正确率，从上一代的 17% 飙升至 56%。 有研究者怀疑，MiniMax 可能在旧版 M3 接受第三方评测时，获取了未公开的测试题目，并用于训练新模型。 该研究者修改了一道公开测试题，让正确答案发生变化，结果 M3.1 Flash 仍然输出原题答案。
This post quotes another post: “Suspicion that Minimax M3.1 Flash benchmaxxed AA Omniscience, passing Gemini 4 Argon”
x.com/gshota666/status/2112574204168327420Archived 2026.10.21 12:05