M3.1 Flash 与 Opus 5.5 同任务成本对比:5M token 任务 $3 vs $100
按官方定价算,M3.1 Flash 输出 $0.60/百万 token,Opus 5.5 为 $20,同一 5M token 任务分别约 $3 和 $100;作者提醒发布跑分(DeepSWE 73.7%、CyberGym 82%)是厂商自评,权重未放出,建议用自己的仓库实测。
A SECRET MODEL ATE 200 TRILLION TOKENS, AND NOW ITS PRICE LOOKS LIKE A TYPO it ran anonymous as "Space Bunny" and chewed through 200 trillion tokens before anyone knew its name today it's MiniMax M3.1 Flash the question everyone keeps asking: "is the cheap model good enough to replace the expensive one?" the better question: how much cheaper is it, per job, in real money? Claude Opus 5.5 reads both price lists and does the sum: → Flash output: $0.60 per million tokens → Opus 5.5 output: $20 per million → divide them... 33 Flash tokens for every single Opus token → a 5M-token job: $3 on Flash, $100 on Opus → the launch chart says 73.7% on DeepSWE and 82% on CyberGym the trap: a launch chart is the lab grading its own homework, and the weights aren't even out yet Flash is the price, Opus is the check, your own repo is the judge the receipt in five lines > the same job, written by both > Flash bills $3, Opus bills $100 > the chart says they're close > your tests say who's right > the winner is whoever passes with the smaller bill go on, send your most boring task through both models tonight and see which bill makes you laugh