Create

Sign in to ReadmeX

or

How we handle your data: Privacy Policy

ReadmeX

Your AI reading companion

What would you like to catch up on?

Pick a briefing, or ask your own question.

Your community and people briefings stay personal to you.

Story

MiniMax M3.1 Flash benchmark gains spark training-data allegations

AI summary

MiniMax M3.1 Flash earned an Artificial Analysis Intelligence Index score of 42, 13 points above M3, but researchers questioned whether the gain reflected genuine capability or memorized benchmark questions. Researcher Binkg reported that the model retained the original answer after a public AA-Omniscience question was modified, and TMTPost reproduced a similar English-language failure, although the model answered correctly in Chinese. MiniMax opened the model weights on October 23, enabling independent testing, but no request logs or training records show that the company accessed or used private evaluation questions.

In-depth insight

The unresolved allegation raises questions about the integrity of third-party model rankings, although the evidence currently supports a need for further testing rather than a finding of benchmark manipulation.

中国人工智能企业MiniMax于2026年10月19日发布了新一代模型M3.1 Flash,声称该模型在智能体任务、编程和网络安全等领域实现了显著进步,同时将在线推理速度提升至上一代的近3倍,并将使用成本降低约70%。然而,模型发布后不久,一项围绕其独立评测成绩的争议迅速引发关注:有研究者怀疑,MiniMax可能在上一代M3模型接受评测时接触到了评测机构Artificial Analysis使用的部分测试问题,并将相关内容用于训练M3.1 Flash,从而使新模型在特定基准测试中获得异常突出的成绩。

需要强调的是,截至相关报道发布时,这仍然是一项基于有限现象提出的指控,尚无公开证据能够证明MiniMax确实窃取或利用了评测题目。相关争议的核心,是人工智能行业长期存在的一个难题:模型在基准测试中的高分,究竟反映了真正的通用能力提升,还是部分来自针对测试本身的优化。

此次争议的关键在于Artificial Analysis的AA-Omniscience评测。这是一项侧重事实知识与回答可靠性的测试,题库总量约为6000道题,其中只有约600道题公开,其余题目并未公开发布。按照常规理解,未公开题目能够降低模型通过专门训练来提高测试成绩的可能性,但如果模型开发商能够在评测过程中接触到这些题目,这种保护机制就可能受到挑战。

#MiniMax#MiniMax M3.1 Flash#MiniMax M3#Artificial Analysis#AA-Omniscience#DeepSeek V4.1 Flash-1017#Gemini 4 Argon

Original · 7 min readRead the original
Group chat

No comments yet. Start the conversation.