**Julian Goldie** (0:00)
China's MiniMax M3 just shocked the AI world, and I want to show you exactly why. I want to run through the benchmarks, I'm gonna show you exactly what I built with this week, because the model was impressive on its own, but what it does inside the system that I just built is absolutely wild. And here's the headline, because this totally changes how you think about business, how it works, et cetera. So on June the 1st, a Chinese company called MiniMax dropped this, a new model called M3, and it's beating the big American names with some serious tests, Gpt 5.5, Gemini 3.1 Pro, uncertain coding tasks, M3 actually edged past both for a tiny little fraction of the price. Now, if you haven't seen them already, these are the benchmarks. And the part that got me right here, they gave M3 one brutally hard job, no answer to copy, just go figure out and actually ran its own for 24 hours straight, a full day, nearly 2000 separate moves, 14, sorry, 147 tries, nobody touched it along the way, 24 hours alone, nobody in the room. So let me show you what that actually means for you. And let me start with a big idea because it literally changes everything. You want to stop thinking of AI as a chatbot and start thinking of it as a team. That's the whole shift. And to show you what I mean, let me show you the old way first because most people are still doing AI the old way without even realizing it. The old way looks like this, one box, you type a question, it answers, then you copy that answer, paste it somewhere else, open up a different tool, do the next bit, copy that, paste it again. You are the glue, you run it between 10 different apps holding it all together with your own hands. And the second you close the tab, it just forgets everything, right? With no single memory. I'll be honest with you, those numbers come from MiniMax's own lab, not outside checkers yet. So take them as a strong first signal, not gospel, but the direction is clear here. And here's the framework I want you to walk away with today. So I call it the night shift. And here's what I mean. Right now, your business is one shift.
You work, the business works, you stop, the business stops, you sleep, nothing happens. And the night shift is when your AI keeps working even after you've logged off. You hand it the job at 6 p.m., you have dinner, you watch a film, you go to bed, you watch Netflix or something, and it's still grinding away at 2 a.m., 4 a.m., 6 a.m. You wake up to finish work. That's exactly what we've done with the studio, as you can see right here. This will work 24-7 if I want it to.
So the difference here is this is an agent operating screen. So it has one screen, all my AI workers lined up together, Claude, OpenClaude, Hermes. They share one memory, so you all know what the other is doing. It's not 10 apps, not 10 logins, and it's just one room where the whole team reports in. This week, I actually added a new worker to that room, which is MiniMax.
So it turned one of my agents into a studio, and it could run itself. Now, I want to show you two big things about M3 that makes it a real night shift worker. First of all, the memory. So M3 actually holds about a million words in its head at once. For example, picture, handing it a stack of books, taller than you, remembers every page whilst it works. The old AI forgot to start the conversation by the end of it. This one holds your whole business at once. You could drop in everything, every note, every past email, your full content folder, and it works off the complete picture, not a tiny scratch.
For example, a coach could hand it a year of client check-ins and ask it to spot which clients are about to drop off. Now, the next thing is the power of this, right? So second, it doesn't just read. M3 was actually built from the ground up to handle pictures and video too. So you can look at a video, it can make an image, it can make a voiceover, and that's the piece that turns one of my agents into a full studio, which I'll show you live in seconds. So now you've got a worker with a giant memory that handles words and pictures and sound. They'll grind it out for a full day without quitting. The only thing left is to give it a body, something that actually goes and does the work whilst you're gone. And that body is Hermes. And right here, you can actually plug it in. So M3 plugs straight into Hermes. The simplest way to see it is like M3 is the brain, Hermes is the hands, the brain decides the hands do, Hermes opens the files, runs the tools, makes the thing and keeps going on its own. So I literally took that brain, dropped it into the hands and parked both inside my agent operating system. Next to the others and that's a night shift worker, fully built. Now let me show you what it does. So inside Hermes, for example, there's one box and you can type what you want and it just goes off and makes it. So for example here, we've generated a image of a dragon. We could, for example, plug this in, just hit generate and then it can go off and create whatever we want. Now, not just that, but it can also use voiceovers and it can generate with videos as well. So as an example of that, we go to another tab. We're going to open up Hermes. You can see, for example, we've created videos using the same sort of process here. And so we can use Hermes workspace to generate images. Look at that image we just created down a couple of clicks. We can go over to video and we could type a video and get that working. We could get a voiceover so we could have a podcast. We can generate anything that we want. And this all works in the background and then we're good to go. Now, if we go to the workspace section here, we've got images, we've got our folders, we've got everything organized in one place, all sorted, all ready to grab. So you don't need to dig through 10 apps for one thing that you made an hour ago. So, for example, like a creator could line up a bunch of posts, like before they go to sleep and then wake up with the whole batch sitting in a folder. That's a night shift, one box, one dashboard and a brain that won't forget and hands up won't quit. Now, here's the part where most people stop themselves. So they look at all of this and they think that's great for you, but I'd never set it up. I'm not techie. I hear it every single day and I'll be straight. Writing this up yourself, building this out yourself, tokens, folders, settings and one wrong stuff and nothing connects, it's not hard. Once you know the order, the problem is nobody hands you the order. That's the whole reason that we actually build the AI Profit Boarding. So we've got over 3,300 business owners inside and loads of them have never opened an AI tool before they opened this. You can see all the examples right here, but these are total beginners. Now inside the AI Profit Boarding, we actually hand you the whole agent operating system, the exact set up with Hermes and MiniMax, Claude already working as a team, the zip file to start with, a 30 day roadmap, the walks you through, plugging in step by step video tutorials for every click, daily updates as I keep improving it, plus four week live coaching calls where you share your screen and someone gets you on stock on the spot. So at 11 p.m. when something won't connect, you're not alone with it. There's always people online. You see right now 116 people online and 3,300 people inside. So if you want to get access to that, link in the comments description or just go to the aiproperborn.com. Let me go back to why this one matters more than the usual monthly AI noise. Most of these models are a bit smarter than the last one. And that's fine. This one's different because of three things stacking at once. Number one, it's frontier level. It was a night shift and it's cheap. So you pick any two and you've never had it before. All three together at this price. That's a shock. And here's what most other AI can't do. The thing buried in that 24 hour test. Most models give up. MiniMax actually said it plainly. Most quit after about 30 tries. MiniMax M3 and its best answer didn't just land until number 145 So it hit wall after wall, just kept plugging away. And the jobs that drain you aren't the quick ones, right? They're the long, boring, monotonous stuff where you just wish you had someone who could just go off and do it for you. You want a worker that could try 100 times and just do it whilst you're not even there. And that's basically what you're getting with MiniMax M3. Now, let's put the whole picture together here. You've got Hermes, which is a brain that holds your entire business and won't forget it. So you've got that's M3. Then you've got the hands, which is Hermes, that doesn't work without you watching. We even got a goal section here so you can just type in a goal and it just gets to work. You got the studio that creates your images, your voiceovers, your videos, et cetera. It's all in one dashboard. And it's not just doing this, but it's sharing one memory together because we've actually plugged in, we've plugged in Obsidian into that. So you've got a really powerful free memory that you can plug in as well. And that's not a chat bot anymore, right? That's your first hire and it works the shift you can't. Now, some of you thinking this is moving too fast or waiting till it comes in. I'll get it. It's exhausting, but it's not going to calm down. The people we know aren't the ones who wait for the dust to settle. They're the ones who started whilst it was all messy because by the time this feels safe, the early movers are already going to be charging more, working less and quietly taking the clients you always wanted. If you're worried like I can't code, you don't need to. Nothing I showed you today needs code. You already get the zip file and you're ready to go with that. Also with using just MiniMax in general, you don't need to code, you just need to tell it what you want. People in our group have never touched AI. A few months ago, I'm running their own night shift right now.
2 more minutes of transcript below
Try it now — copy, paste, done:
curl -H "x-api-key: pt_demo" \
https://spoken.md/transcripts/1000651996090
Works with Claude, ChatGPT, Cursor, and any agent that makes HTTP calls.
From $0.10 per transcript. No subscription. Credits never expire.
Using your own key:
curl -H "x-api-key: YOUR_KEY" \
https://spoken.md/transcripts/1000771137650