MiniMax M2.5发布
Xin Lang Cai Jing·2026-02-12 23:43
Core Insights - The MiniMax M2.5 has been released, enhancing its ability to decompose complex tasks and optimize token consumption during the thought process [1] - In SWE-Bench Verified tests, M2.5 completed tasks 37% faster than its predecessor, M2.1 [1] Performance Metrics - At a token output rate of 100 tokens per second, the cost for continuous operation for one hour is $1 [1] - At a token output rate of 50 tokens per second, the cost for continuous operation for one hour is $0.3 [1] - A budget of $10 million can sustain 4 agents working continuously for one year [1]