MiniMax M3.1 Flash 官方规格：$0.15/$0.60，200+ TPS
官方发布信息：agent 任务成本比前代低 70%，输入 $0.15/百万、输出 $0.60、缓存输入 $0.015；自报 DeepSWE v1.1 73.7%、OSWorld V2 61.7、CyberGym 82%，P50 170+ TPS，权重稍后开源。
MiniMax released M3.1 Flash on 19 October 2026 and says overall usage cost for agentic tasks is 70% lower than its predecessor: $0.15/1M input, $0.60/1M output, $0.015 cached input. MiniMax reports 73.7% on DeepSWE v1.1, 76.2 points on BankerToolBench, 61.7 points on OSWorld V2, 82% on CyberGym. Speed: output can reach 200+ TPS, median (P50) consistently 170+ TPS, which MiniMax credits to more aggressive sparsity. An early version ran anonymously as Space Bunny and processed more than 200 trillion tokens. All figures are MiniMax's own reported results; weights ship later. If you run agent fleets, what task would you benchmark it on first?