M3.1 Flash 官方规格：$0.60/M 输出、1M 上下文、10/23 开源权重

Vaibhav Sisinty @VaibhavSisinty原推发布于 2026.10.20

在 X 上打开原推

$0.60 per million output tokens. That is the real MiniMax headline M3.1 Flash is a multimodal coding model with a 1M token context and tunable thinking depth MiniMax scores it at 62.2 on Automation-Bench and 91.0 on BabyVision, though DeepSeek still leads DeepSWE 77.0 to 73.7 At this price, agents can run longer, retry mistakes and verify their own work instead of stopping after one expensive attempt MiniMax says it costs one third of M3 and runs three times faster The API is live now, with weights coming on October 23

这条是引用着另一条写的：x.com/1875078099538423808/status/211215086786854…