WHAT NEW IN X10.29 THU

AI collections from X · 22 collections · 58,925 posts

minimax-m3.1POST

MiniMax M3.1 Flash launch specs: one-third the cost of M3, 3x faster, open weights on Oct 23

A rundown of M3.1 Flash (formerly Space Bunny): $0.15/$0.60 per million tokens, Automation-Bench 62.2%, LOCA-Bench 256K 60.3%, with HLE 43.4% and Terminal Bench 38.1% flagged as weak spots.

AI超元域
@AISuperDomainArchived 8 days agoOriginally posted 2026.10.19
AI超元域 @AISuperDomainOriginally posted 2026.10.19
Open original on X

跑多 Agent 编排和编码,现在输入成本压到了 $0.15 / 百万 tokens。 MiniMax 刚发布 MiniMax M3.1 Flash(即此前吞吐量超 200T tokens 的 Space Bunny) 相比 M3,成本降为 1/3、提速 3 倍,专为高频调用工具与并行跑 Subagents 优化: • Agent 与长上下文第一:Automation-Bench 62.2%、256K 长上下文 LOCA-Bench 60.3%、工具调用 BankerToolBench 76.2%,均列官方对比表榜首 • 编码与视觉扎实:DeepSWE 达 73.7%(逼近 Opus 5 的 74.0%),BabyVision 达 91.0% • 定价与开源:输入/输出为 $0.15 / $0.60 每百万 tokens(缓存读取 $0.015);10 月 23 日开源权重 客观短板:无工具极限推理(HLE 43.4%)及命令行任务(Terminal Bench 38.1%)距 Opus 5 等重型旗舰仍有差距,更适合做高频 Agent 与日常编码的主力执行模型

This post quotes another post: x.com/minimax_ai/status/2112150867868541221

x.com/aisuperdomain/status/2112157153662259476Archived 2026.10.20 20:25