WHAT NEW IN X10.29 THU

X 上的 AI 合集档案 · 22 个合集 · 58,925 条

minimax-m3.1POST

MiniMax M3.1 Flash 官方规格:跑分、定价与 10/23 开源权重

官方数据:Automation-Bench 62.2 超 Opus 5 的 50.3,但 OSWorld 与 Terminal Bench 落后;API $0.15/$0.60,10 月 23 日放权重,并确认 Space Bunny 是早期版本。

Fokki
@0x_fokki8 天前收录原推发布于 2026.10.19
Fokki @0x_fokki原推发布于 2026.10.19
在 X 上打开原推

🚨 IN 4 DAYS YOU CAN DOWNLOAD A MODEL THAT OUTSCORES OPUS 5 ON AUTOMATION-BENCH MiniMax M3.1 Flash. You can call the API today and download the weights on Oct 23. > Automation-Bench: 62.2 vs Opus 5's 50.3 > LOCA-Bench 256K: 60.3 vs Opus 5's 53.7 > API: $0.15 input / $0.60 output per 1M tokens > Against M3: 3x the speed at 1/3 the cost Opus 5 still wins OSWorld 70.6 to 61.7 and Terminal Bench 51.8 to 38.1. And Space Bunny, which topped leaderboards with 200T+ tokens? MiniMax says it was an earlier version of M3.1 Flash. looks good on paper. I want to run it on my own stuff when the weights drop.

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

x.com/0x_fokki/status/2112173234921656515采集于 2026.10.20 20:27