01What Is Minimax M3 and How It Works

Minimax M3 is a new AI model from a Chinese company called Minimax, released on June 1st. It is a frontier-level model that beats big American names like GPT-5.5 and Gemini 3.1 Pro on certain coding tasks, and it costs a fraction of the price.

How does it work? M3 is built to handle not just text but also pictures, video, and voiceovers. It has a huge memory that can hold about a million words at once, so it remembers everything you give it while it works.

In a 24-hour test, M3 ran on its own for a full day, making 147 tries without anyone touching it. Its best answer didn't come until try 145, showing it keeps going when other models give up.

The big idea is to stop thinking of AI as a chatbot and start thinking of it as a team. M3 is the brain, and when you plug it into a system like Hermes, Hermes becomes the hands that do the work. That combo turns your AI into a 'night shift' worker that keeps grinding while you sleep.

02Main Features of Minimax M3: Memory, Media, and Persistence

M3 has three main features that make it special. First, its giant memory holds about a million words at once. That's like handing it a stack of books taller than you, and it remembers every page.

You can drop in all your notes, past emails, and content folders, and it works off the complete picture, not a tiny scratch.

Second, M3 is multimodal—it can look at video, make images, and create voiceovers. That turns it into a full studio, not just a text bot. Third, it has persistence.

In the 24-hour test, it made 147 tries without quitting. Most models quit after about 30 tries, but M3 keeps plugging away until it finds an answer.

Here is a list of what M3 can do:

  • Hold about a million words in memory at once.
  • Generate images, videos, and voiceovers.
  • Read and understand video content.
  • Keep working on a task for 24 hours straight.
  • Make over 140 attempts to solve a hard problem.

These features stack up to make M3 a real 'night shift' worker that never forgets and never quits.

03How to Set Up Minimax M3 with Hermes

You don't need to be a techie or know how to code. Here's the simple order to get M3 working inside an agent operating system like Hermes.

  1. Sign up for an agent operating system like Hermes, or get the AI Profit Boardroom's zip file that has everything pre-set.
  2. Plug M3 into Hermes. Think of M3 as the brain and Hermes as the hands—the brain decides, the hands do.
  3. Add M3 to your agent operating room alongside other workers like Claude and OpenClaude.
  4. Give all your agents one shared memory so they know what each other is doing.
  5. Optional: Plug in Obsidian as a free memory base to store notes and files.
  6. Start typing what you want in the one box, and let it go off and make it.

In the AI Profit Boardroom, you get the exact setup already done. They hand you a zip file, a 30-day roadmap, and video tutorials for every click. At 11 p.m. when something won't connect, there are always people online to help you get unstuck.

04How to Use Minimax M3: Examples from the Episode

In the episode, Julian showed how to use M3 inside Hermes. He generated an image of a dragon in a couple of clicks. Then he opened another tab to create videos using the same process.

He also created voiceovers for a podcast. All of it works in the background while you do other things.

Here's the step-by-step from the demo:

  1. Open Hermes and find the one box where you type what you want.
  2. Type something like 'generate an image of a dragon' and hit generate.
  3. Go to the video tab and type a video prompt to get a video working.
  4. Create a voiceover for a podcast or any other audio need.
  5. Check the workspace section to see all your images, videos, and folders organized in one place.

The workspace keeps everything sorted, so you don't dig through 10 apps for something you made an hour ago. You can also use the goal section—just type a goal and it gets to work on its own. For example, a creator could line up a bunch of posts before sleeping and wake up to the whole batch sitting in a folder.

05Results and Comparisons: M3 vs GPT-5.5 and Gemini

On certain coding tasks, M3 edged past GPT-5.5 and Gemini 3.1 Pro for a tiny fraction of the price. The numbers come from Minimax's own lab, not outside checkers yet, so Julian called them a 'strong first signal, not gospel.' But the direction is clear.

The 24-hour test was the most impressive part. M3 was given a hard job with no answer to copy and had to figure it out on its own. It ran for 24 hours straight, made 147 tries, and nobody touched it.

Most models quit after about 30 tries, but M3 kept going and landed its best answer at try 145.

Here's a comparison based on the episode:

  • Coding tasks: M3 beat GPT-5.5 and Gemini 3.1 Pro.
  • Persistence: M3 made 147 tries vs. most models quitting after 30.
  • Best answer: M3's best came at try 145, after many failed attempts.
  • Price: M3 costs a 'tiny fraction' of the big American models.

Julian also showed the old way vs. the new way. The old way is one box, you type, copy, paste, and you are the glue. The new way is an agent operating screen where all AI workers line up, share one memory, and report to one room.

06Tips, Limits, and Mistakes to Avoid with Minimax M3

Here are some tips and warnings from the episode. First, don't take the benchmarks as gospel. They come from Minimax's own lab, not outside checkers yet.

Treat them as a strong first signal, but wait for independent tests if you want proof.

Don't try to build this yourself from scratch. Julian said writing it up yourself involves tokens, folders, settings, and one wrong step and nothing connects. It's not hard once you know the order, but nobody hands you the order. That's why he built the AI Profit Boardroom with 3,300+ business owners inside, many who have never opened an AI tool before.

Avoid thinking of AI as just a chatbot. That's the old way. You end up being the glue between 10 different apps, and the AI forgets everything when you close the tab.

Instead, use an agent operating system with shared memory so your AI works as a team.

Also, don't wait for the dust to settle. Julian said it's exhausting, but AI isn't going to calm down. The people who win are the ones who start while it's messy.

By the time it feels safe, early movers will be charging more, working less, and taking the clients you wanted.

07Who Is Minimax M3 For and Why It Matters

Minimax M3 is for anyone who runs a business and wants to stop being the only worker. It's for coaches, creators, and business owners who are tired of doing every shift alone. Julian showed how a coach could hand M3 a year of client check-ins and ask it to spot which clients are about to drop off.

It matters because it's the first time frontier-level AI is cheap enough and stubborn enough to work all night. The jobs that drain you aren't the quick ones—they're the long, boring, monotonous stuff where you wish someone could just go off and do it. M3 can try a hundred times and keep going while you aren't even there.

You don't need to code. Julian said nothing he showed requires code. You just tell it what you want.

People in his group who had never touched AI a few months ago are now running their own night shift. The AI Profit Boardroom gives you the whole agent operating system with M3 and Hermes plugged in, plus coaching calls and a community of 3,300 members.

The only question left is who builds their night shift now and who does every shift alone.