Space Bunny(M3.1 Flash 匿名版)揭晓:自动化 62.2%,价格仅 Opus 5 的 1/30
作者确认 Space Bunny 就是 MiniMax,M3.1 Flash 自动化跑分 62.2% 超 Opus 5 与 Sonnet 5.6,DeepSWE 持平,200+ tok/s,$0.15/$0.60 对比 Opus 5 的 $5/$25。
so Space Bunny was from MiniMax all along the official version M3.1 Flash scored 62.2% on automation-bench that's ahead of both Opus 5 and Sonnet 5.6, which honestly I didn't expect from a Flash model it matches them on DeepSWE too, and runs at 200+ tokens/sec but the pricing is the part I care about most $0.15 / $0.60 per million tokens while Opus 5 is $5 / $25 for agents that run all day, that gap matters way more than a few benchmark points I knew Space Bunny was too good to be a random model MiniMax is quietly becoming my favorite lab to watch
这条是引用着另一条写的:x.com/minimax_ai/status/2112150867868541221