MiniMax M3.1 Flash open weights: 428B/23B active, 1M context, vendor-vs-outside benchmark gap, $20M license line
An article on M3.1 Flash's open weights covering 428B total / 23B active parameters, 1M context, vendor benchmarks running higher than outside numbers, a $20M clause in the license, and what to check before self-hosting.
What this article says
文章围绕 MiniMax 公开 M3.1 Flash 权重这件事，给出自托管前的核查框架。模型规格为 428B 总参数、每 token 激活 23B、支持 1M 上下文。作者指出一个关键落差：厂商自己公布的 benchmark 成绩相当亮眼，但外部独立测量的数字普遍更低，因此选型时不能只看官方表格。许可证方面存在一条 2000 万美元量级的条款，属于商用或规模门槛，需要提前确认自己的使用场景是否触发。文章的核心主张是：权重开放不等于可以直接用，团队在决定自托管前应逐项核对参数量与显存需求、量化方案、推理引擎支持、许可证边界，以及外部实测与官方跑分的差距，再判断自建是否真的比调 API 划算。
MiniMax just made M3.1 Flash's weights public: 428B parameters, 23B active, 1M context. The vendor benchmarks look strong, the outside numbers are lower, and the license has a $20M line. What to check before you self-host. https://x.com/i/article/2113681381561888768