MiniMax-M2: New Open-Source Model That's 2x FASTER Than Claude!

So, there's this new model that's kind of blowing up right now, and it's called Miniax M2. And people are calling it the fastest open-source model ever built for coding and agents. It's open, it's cheap, and it's ridiculously quick. In the wild part, Miniax is claiming it runs at twice the speed of Claude Sonnet for about 8% of the cost. Yeah, that's not a typo. So, in this video, let's talk about what M2 actually is, why it's getting so much attention, and what makes it different from every other new model we've seen this year. So, let's

start with the company. They're a global AI foundation model company founded early in 2022, and their mission is intelligence with everyone, which basically means making powerful AI tools accessible to anyone and everywhere. So, they really tap into that open-source model. Their goal is to push the frontiers of AI towards AGI. And they're not just building text models. They've actually developed a full suite of multimodal systems like Halo 2. 3 for conversation, speech 2. 6 for voice, and music 2. 0 for audio generation. All of these models, including the M21, can understand and create across text, audio, images, video, and even music. Their ecosystem already powers a bunch of products you might have seen floating around. Miniax Agent, Helio AI, Talki, Miniax Audio, and their Open AI platform for developers. So far, their models have reached over 157 million users across more than 200 countries, and over 50,000 companies and developers use their tech globally. That's a pretty serious footprint for a company that most people in the West haven't even heard of yet. So, what exactly is

Miniaax M2? This is actually their latest flagship foundation model and it's what they call a mixture of expert system ore meaning it's huge but smart about how it works. It has 230 billion total parameters but only 10 billion activate during inference. So instead of using the full brain or full power every single time it activates just the right experts for the job. It's kind of like calling in a specialist team rather than you know calling the whole entire company. That setup alone gives the model its cost effectiveness. It gives it big model power, but with a small model efficiency, and that's what makes it so fast and affordable to run as well. Miniax M2 at the moment is free for a limited time. So, if you're interested in trying this model out, you can click the link in my description.

You know what's interesting? Miniax actually didn't just build this model out of nowhere. They actually started by building agents, real in-house AI assistants to help their own teams handle the chaos of growing so fast. At first, these agents were pretty simple. But over time, they started taking on bigger and bigger jobs, researching technical problems, analyzing data, writing code, processing user feedback, and even screening resumes for HR. Basically, these agents became part of the company. They kind of worked alongside the humans helping Miniax become what they call an AI native organization and that's where the team realized something important. They believe AGI isn't just intelligence, it's production power. It's a new kind of workforce and agents are how that workforce takes shape evolving from simple chat bots into real digital teammates that can also think, plan and complete tasks independently. But there was a problem. No single model at the moment, according to them, could handle what they needed. The fast ones were too expensive. The cheap ones weren't smart enough. It was what they called the impossible triangle. Performance, price, and speed. You could pick two, but never all three. They saw a huge gap in the market and not just for themselves. Even the big companies trying to build agent systems were paying hundreds of dollars per user or waiting hours for a single task to finish. So the team asked, what if we could break that triangle? What if they could build a model that was smart enough to code, reason, and use tools, but fast and cheap enough for anyone to actually deploy? That question, that mission is what led them to build Miniax M2. A model designed not just for speed or scale, but for real work. That kind of AI makes the agent era accessible to everyone and that also allowed them to carry forward their vision of intelligence with everyone and is not

just fats is actually really good on artificial analysis global rankings. Miniax M2 is currently number one among all open- source models for overall intelligence. It scored incredibly well across math, science, instruction following, coding and tool use. A few numbers worth mentioning are that the software engineering bench verified it scored 69. 4 basically on par with GPT5 and claude 4. 5. On the terminal benchmark 46. 3 great at executing and repairing code in shell environments. On the browse comp benchmark 44 which is a huge jump from other open models when it comes to online search and retrieval. in the fin search comp benchmark 65. 5 which means strong in financial and data reasoning. Miniax solves the impossible triangle problem because it allows you to select performance, speed, and price with this model. All right, so let's

test out this model for real. And first thing you can see is that I have selected the pro mode. You also have the ability to change your MCP. There's some other settings you can play around with like environment variables, whatnot. Um, so the pro mode is actually available for free for a limited time and you guys have access to this as well. So feel free to try it out. And at the bottom you can see that there is kind of like a suggestion of projects that people have done. You can see that there's coding projects, there are some research projects or more. So I haven't checked them out, but uh feel free to check those out and see what people have created. But for today's video, what we're going to do is we're going to ask Miniax to create a Tetris game. So, let's give it that. All right. Let's press run. So, it's running right now. Um, okay. So, it's starting to execute the quote or more like it's coming up with a game plan of what it wants to do. And you can see like the game plan is quite transparent. So it's almost like you're working with a employee and it's telling you how they're going to produce the uh project at hand. Um so yeah there's some good details here about it. Um and then okay what is it doing now? Now it's actually checking the memory to see if we've actually given this prompt in the past. If it has some you know past memory or data that it can use it's going to utilize that. So that's kind of how it operates to be more efficient. But we can see that obviously this is the first time we're using this model. So there is a no memory. Now it's asking me do I want to design it as a classic Tetris game or like custom design it. I'll just decide to do like the classic Tetris version. Um so let's give it that. So it's executing. Yeah, it's also thinking of key features about the website. So you can see it's like very um detailed and plan oriented. You can also see the process is on the side of the screen. So if you're trying to execute and trying to see what's happening, you can also see it on the side. And uh so let's see what it's doing. All right. So it looks like we have a version of Tetris. So we can see that the project is actually done being built. But what's cool about Miniax 2 is that it's actually going to test the project for us. So it's going to come with a plan to test it. So, it has come up with about 11 testing areas that it wants to test the code on, which is I think is cool because not only is this agent like creating something for you, but it's also thinking about how it can better itself. So, it's going to come up with a testing plan that we can see on the side. Um, and you can see it has like a checklist kind of and it also has the steps that it wants to test everything in. So, that's pretty cool. I think that's really important for us, especially when you're creating something because, you know, AIS can make mistakes sometimes. But if you have an AI that's able to clean itself, test itself, and improve, why not, right? So, we can see it did some tests. Uh, it did fail something. So, that's cool. It also recognizes what has failed. It's going to run it again and try to fix that failure as well. So, that's going to be good to see. So, we can see that it's like an ongoing process. It tests itself, it fixes itself, it breaks something, and then it fixes itself. So which is really cool. So let's let it run a little bit more. So more testing is going on at the moment. Um you can see it has gone through multiple repetitions of the test which so it's trying to improve the end product which is really important. All right. You can also see that all the files that it has been creating. So for example the screenshots is used for its testing are over here. Um you can see it recognized something was wrong. So it has those screenshots in there. You also have the memory all of that here. You also have the whole game file like all the code for it. You also have the test progress. So as I created like a kind of like a progress for the testing process over here as well. Okay, so it looks pretty good. Looks pretty clean. Let's start the game. Um, you know, the buttons seem to be working. Oh, the space button kind of seems to be glitching a little. Every time I try to press hard down, it doesn't really do it. As you can see, I tried it again. But, you know, the game works. Let's see if it's actually able to do the point system correctly. So, if I get a line out, does it give me points for it? So, let me play some Tetris. I'm not a pro Tetris player, but I wouldn't say I'm bad. So, let's see. Okay. It does give us this. This is cool, guys. Like, I just literally gave it a prompt. Make me a Tetris. I haven't said much and I said classical Tetris and he has come up with something that I can actually interact with, play

and use. So, it's crazy. So, that's Miniax M2, the model built to make AI agents faster, smarter, and accessible to everyone. You can click the link in the description to try it out yourself. And if you enjoyed this breakdown, make sure to like, subscribe, and turn on notifications. There's a lot more coming your way soon. See you in the next