WHAT NEW IN X10.29 THU

X 上的 AI 合集档案 · 22 个合集 · 60,636 条

minimax-m3.1IMAGE

实测规划:把 M3.1 Flash 当 Opus 5.5 的执行层,$0.15/$0.60 每百万 token

作者列出本周三处接入点:30 步抓取循环的执行器、PR 初审、夜间研究批处理,并引用 BankerToolBench 76.2、DeepSWE 73.7、Automation-Bench 62.2 与 170+ tok/s。

Mr. Buzzoni
@Mr_Buzzoni8 天前收录原推发布于 2026.10.20
实测规划:把 M3.1 Flash 当 Opus 5.5 的执行层,$0.15/$0.60 每百万 token
Mr. Buzzoni @Mr_Buzzoni原推发布于 2026.10.20
在 X 上打开原推

MiniMax M3.1 Flash is the Flash-tier model I've been hunting for to run under Opus 5.5 without blowing the agent bill the three spots I'm wiring it into this week: > executor on my 30-step scraping loop 76.2 on BankerToolBench, holds the thread across files > first pass on PRs before I pay Opus to review 73.7 on DeepSWE, right next to the frontier > overnight worker in my research vault, chews through 40 pages while I sleep 62.2 on Automation-Bench, ahead of everyone in the chart at $0.15 / $0.60 per 1M and 170+ tokens a second, a weekend-long agent run still fits in a lunch tab turns out Space Bunny - the stealth model that's been eating leaderboards - was M3.1 Flash the whole time

这条是引用着另一条写的:x.com/minimax_ai/status/2112150867868541221

x.com/mr_buzzoni/status/2112227526802120731采集于 2026.10.20 20:14