Skyler Miao (@skylermiao7) 's Twitter Profile
Skyler Miao

@skylermiao7

Head of Engineering @MiniMax_AI
Building MiniMax M2.x, Agent, Audio and @Hailuo_AI

ID: 1879314108723658752

linkhttp://minimax.io calendar_today14-01-2025 23:46:30

890 Tweet

5,5K Takipçi

276 Takip Edilen

kevinma (@kevinma_dev_zh) 's Twitter Profile Photo

今天被 MiniMax M2.5 刷屏,这东西结合 OpenClaw 确实很可能改变了游戏规则。 简单说,就是把一个性能堪比顶级闭源模型(编程能力超 Opus 4.6)、但几乎免费的开源大脑(M2.5),装进了一个能控制你电脑的开源身体(OpenClaw)里。

Skyler Miao (@skylermiao7) 's Twitter Profile Photo

This is one of the most real engineer-angle tests I’ve seen on M2.5, very similar to what we run internally, and we’ve even turned these into automated benchmarks. Building M-series as a real engineering teammate on real tasks has always been the goal. Still a lot of headroom,

MiniMax (official) (@minimax__ai) 's Twitter Profile Photo

hi elie, thanks for your question > 10B active parameters was intentional > M2.5 is getting close to “infinite agent scaling” > knowledge capacity is the main limit > tradeoffs are thoughtfully chosen for efficiency & practicality > pretraining innovations remain exciting areas

hi elie, thanks for your question 
> 10B active parameters was intentional
> M2.5 is getting close to “infinite agent scaling”
> knowledge capacity is the main limit
> tradeoffs are thoughtfully chosen for efficiency & practicality
> pretraining innovations remain exciting areas
Skyler Miao (@skylermiao7) 's Twitter Profile Photo

Appreciate it. The gap is closing fast. What comes next isn’t just higher scores — it’s optimisation, efficiency, real-world agents. Intelligence cost → trending to zero. Tokens will feel like water and electricity. That day is coming.

MiniMax_Agent (@minimaxagent) 's Twitter Profile Photo

Valentine's romance with M2.5 and MiniMax Agent ❤️♟️ Write to the one you love, whether they’re near or far. y1amjtt2fyvc.space.minimax.io

Ahmad (@theahmadosman) 's Twitter Profile Photo

MiniMax-M2.5 NVFP4 running on 2x RTX 6000 > 83 tokens/sec single stream > allows up to 32 concurrent users > pushing 1000+ tokens/sec > this costs ~2x 512GB Mac Studios > except Mac = 1 prompt at a time > and even slower tokens/sec > model on HF > lukealonso/MiniMax-M2.5-NVFP4

MiniMax-M2.5 NVFP4
running on 2x RTX 6000

> 83 tokens/sec single stream
> allows up to 32 concurrent users
> pushing 1000+ tokens/sec

> this costs ~2x 512GB Mac Studios
> except Mac = 1 prompt at a time
> and even slower tokens/sec

> model on HF
> lukealonso/MiniMax-M2.5-NVFP4
Build That Idea (@buildthatidea) 's Twitter Profile Photo

Opus 4.6 vs. Minimax M2.5 Prompt: Build an interactive solar system from scratch. Opus 4.6 tried to create a beautiful UI, but the sun’s shadow ruined it. Minimax M2.5 built a simple version that works beautifully. The winner is: 🥇 Minimax M2.5 🥈 Opus 4.6