# These 3 NEW Chinese Autonomous AI Agents are INSANE…

## Метаданные

- **Канал:** Julian Goldie SEO
- **YouTube:** https://www.youtube.com/watch?v=KAOnu-rHfUw
- **Дата:** 21.06.2025
- **Длительность:** 25:48
- **Просмотры:** 110,513

## Описание

Want to get more customers, make more profit & save 100s of hours with AI? https://go.juliangoldie.com/ai-profit-boardroom

Free AI Community here 👉 https://www.skool.com/ai-seo-with-julian-goldie-1553

🚀 Get a FREE SEO strategy Session + Discount Now: https://go.juliangoldie.com/strategy-session

🤯  Want more money, traffic and sales from SEO? Join the SEO Elite Circle👇
https://go.juliangoldie.com/register

🤖 Need AI Automation Services? Book an AI Discovery Session Here: https://juliangoldieaiautomation.com/

Click below for FREE access to  ✅ 50 FREE AI SEO TOOLS 🔥 200+ AI SEO Prompts! 📈 FREE AI SEO COMMUNITY with 2,000 SEOs ! 🚀 Free AI SEO Course 🏆 Plus TODAY's Video NOTES...
https://go.juliangoldie.com/chat-gpt-prompts

- Want a Custom GPT built? Order here: https://kwnyzkju.manus.space/
- Join our FREE AI SEO Accelerator here: https://www.facebook.com/groups/aiseomastermind
- Need consulting? Book a call with us here: https://link.juliangoldie.com/widget/bookings/seo-gameplanesov12

## Содержание

### [0:00](https://www.youtube.com/watch?v=KAOnu-rHfUw) Segment 1 (00:00 - 05:00)

Today, I'm going to show you three brand new Chinese AI agents that just dropped, and they're absolutely destroying everything else on the planet. We're talking about AI that builds Netflix clones in one click, fixes code faster than human developers, and creates Hollywood quality videos in under a minute. And the craziest part, they're all completely free while everyone else is charging you hundreds of dollars a month. I tested all three myself, and what I found will blow your mind. Hey, if we haven't met already, I'm the digital avatar of Julian Goldie, CEO of SEO agency Goldie Agency. Whilst he's helping clients get more leads and customers, I'm here to help you get the latest AI updates. Julian Goldie reads every comment, so make sure you comment below. Look, I've been tracking AI for years now. I've seen chat GPT explode. I've watched every major AI release. But what just happened in China over the past few weeks is different. These aren't just incremental improvements. These are complete game changers that could put entire industries out of business overnight. And here's what nobody's talking about. While American AI companies are charging massive subscription fees and limiting access, these Chinese companies just dropped three tools that do everything better for free. The first one builds websites that look like they cost $50,000 to make. The second one writes code better than senior developers making six figures. The third one creates videos so good you can't tell they're made by AI. I'm going to show you exactly how each one works, how to get access before everyone else finds out, and the specific ways you can use these to absolutely dominate your competition. Cuz what I'm about to show you isn't just impressive, it's business changing. So, Miniax M1 just dropped and this thing is absolutely insane. I mean, I've been testing AI tools for years now and I've never seen anything quite like this. Picture this. I literally typed create a Netflix clone with playable trailers and boom, in about 60 seconds, I had a fully working Netflix replica with actual video trailers that play, not some halfbaked prototype. A real functioning website that looks exactly like Netflix. And here's the crazy part. It didn't cost me a single penny. Now, you might be thinking, Julian, come on. Another AI tool that promises the world. But hold up. This isn't just another chat GPT wannabe. This is something completely different. So, here's the backstory. Miniax is actually a Shanghai based AI startup and they just dropped something that's making waves in the AI world. They're backed by heavy hitters like Alaba, Tencent, and IDG Capital. But what makes this different from everything else out there? Well, first off, it's the world's first openweight large scale hybrid attention reasoning model. I know that sounds like tech jargon, but stick with me. Here's what this means for you and me. This thing can handle 1 million tokens of input. That's like feeding it an entire book series and having it understand every single word. Compare that to Deepseek R1, which can only handle 128,000 tokens. That's eight times less. And the output, it can generate up to 80,000 tokens in a single response. That's like writing a small novel in one go. But here's where it gets really wild. Remember how I said I built that Netflix clone? Well, I wasn't exaggerating. The design, the images, the trailers, everything was generated automatically. And it didn't just create a static page. It created a fully functioning website with playable video content. Now, you might be wondering, Julian, how much did this cost you? Here's the kicker. The entire model was trained for just 534,700. Now, I know that sounds like a lot, but listen to this. Uh, Deep Seek trained their R1 model for 5 to6 million, and Open AI's GPT4 costs over $100 million to train. That's insane efficiency right there. And the best part, it's actually open source under an Apache license. That means you can use it for commercial applications, modify it however you want without any restrictions or payments. So, how do you actually use this thing? It's stupid simple. You just go to minimax. io and you can start using their agent right away. When I first logged in, I got 1,000 free credits. And from what I've seen, that's more than enough to build some seriously cool stuff. Let me walk you through exactly what I did. I typed in create a Netflix clone with playable trailers and hit enter. That's it. No coding, no design work, no video editing. The AI went off and started doing its thing. It scraped images from the web, generated the layout, created the navigation, and even embedded video players with custom AI generated trailers. But it's not just about building Netflix clones. Let me show you some other crazy examples I've seen people build with this. Someone created Tom Riddle's diary from Harry Potter. You know, the one where you write in it and it writes back. Yeah, they built a fully interactive version of that. Another person built a real-time dashboard tracking the Israel Iran conflict. It pulls data from multiple sources, updates in real time, and presents everything in a clean, professional interface. There's even a guy who built a complete Pokédex with all the Pokémon, their stats, animations, and everything. Click on any Pokémon, and you get all the information about them, just like in the games. And

### [5:00](https://www.youtube.com/watch?v=KAOnu-rHfUw&t=300s) Segment 2 (05:00 - 10:00)

here's something really practical. Someone asked it to find the best location for their new office. The AI went off, researched different areas, analyzed costs, accessibility, transport links, and came back with a full report on the best locations. Now, let's talk about the tech behind this for a second. Minia M1 uses something called a hybrid mixture of experts architecture combined with a lightning attention mechanism. What does that mean in simple terms? When doing deep reasoning with 80,000 tokens, only needs about 30% of the computing power that DeepCar 1 would need. At 100k tokens, it uses just 25% of the computing resources. That's like having a sports car that goes twice as fast, but uses half the fuel. The model has 456 billion parameters total, but only activates 45. 9 billion per token. This selective activation is what makes it so efficient. But here's what really blew my mind when I was teasing it. The speed. When I compared it to other tools like GenSpark, Miniax was generating complete websites while GenSpark was still thinking about the prompt. And it's not just fast, it's smart. It was trained using reinforcement learning on everything from mathematical reasoning to real world software engineering. On software engineering benchmarks, it achieved 56% accuracy on swbench verified, dramatically outperforming other models. Let me give you another example. I asked it to create a series of cinematic videos about a Yeti doing a vlog. Not only did it understand the concept, but it generated the prompts, created the videos, and stitched them together into a longer video. The AI even uses something called browser use, where it can actually browse the web, look at websites, and use that information to build better outputs. When I asked it to research my YouTube channel, and build a landing page, it went to my channel, analyzed my content, looked at my LinkedIn, and created a custom landing page based on what it found. Now, there are two versions of this model, Miniax M140K and Miniax M180K, referring to their thinking budgets or output lengths. The 80K version is more powerful for complex tasks, but the 40K version is perfectly fine for most use cases. If you want to use the API, it costs 0. 55 per million input tokens and 2. 20 per million output tokens for the ATK model. That's incredibly cheap compared to other models with similar capabilities. Here's something else that's really cool. The model comes with built-in capabilities for online search, video and image generation, speech synthesis, and voice cloning. And it's not just a text model. Uh it's a complete AI agent that can do multiple things. And the benchmarks, this thing is competing with the big boys. It's competitive with Open AIO3, Gemini 2. 5 Pro, Claude 4 Opus, and Deepse R1 on various benchmarks. Sometimes it beats them, sometimes it's slightly behind, but it's always in the same league. On long context tasks, it absolutely destroys the competition. It got 73. 4% on OpenAI MRCR compared to just 27. 7% for other models. That's because of its massive 1 million token context window. Now, I know what some of you might be thinking. This sounds too good to be true. And I get it. I was skeptical, too. But here's the thing. The code is completely open source. You can download it from GitHub or HuggingFace right now. You can verify everything yourself. They even provide deployment guides for VLM and transformers, so you can run it on your own servers if you want. A server with eight H800 GPUs can process up to 2 million tokens, while 8H20 GPUs can handle up to 5 million tokens. But for most of us, just using their web interface is more than enough. You don't need any technical knowledge. You don't need to know how to code. You just type what you want and it builds it. I've been using AI tools for years now, and this is genuinely one of the most impressive things I've seen. The combination of being able to build complete functional websites with just a text prompt, the massive context window, and the fact that it's free to start, it's a gamecher. Think about what this means for businesses. You could prototype an entire web application in minutes instead of weeks. You could create interactive demos for clients without hiring developers. You could build internal tools and dashboards without any coding knowledge. And for content creators like me, this is gold. I can create interactive experiences for my audience, build custom tools for my courses, and prototype ideas faster than ever before. They used a new training method called CISPO that's way more efficient than traditional methods. The entire reinforcement learning phase used only 512 Nvidia H800's for 3 weeks. That's incredibly efficient for a model of this size. The fact that they could train something this powerful for under 600 car when competitors are spending tens of millions, that tells you something about the innovation happening here. And remember, this is just the beginning. Miniax is calling this Miniax week, and they're planning to release more products. If this is what they're starting with, I can't wait to see what's coming next. But what Bite Dance just released with Seance 1. 0 is on another level entirely. And most people have no idea this even exists yet.

### [10:00](https://www.youtube.com/watch?v=KAOnu-rHfUw&t=600s) Segment 3 (10:00 - 15:00)

Here's what happened. While everyone was talking about OpenAI's Sora, Bite Dance was quietly building something that would make Sora look like a toy. They just dropped Sea Dance 1. 0, and it's not just better than everything else out there. It's not even close. I'm talking about a video generator that can create Hollywood quality videos from just text. But here's the kicker. It doesn't take hours like the old tools. It doesn't even take minutes. This thing spits out 5-second videos at full 1080p resolution in just 41 seconds on a single graphics card. But that's not even the crazy part. The crazy part is how good these videos actually look. We're talking about videos so realistic you can't tell they're AI generated videos with perfect motion, perfect lighting, perfect everything. And before you ask, no, this isn't some overhyped demo that only works in perfect conditions. This thing just ranked number one on both major video generation leaderboards. It beat Google's V3. It crushed Runways Gen 4. It made Open AI's Sora look like it was from 2020. But here's what really gets me excited. This isn't just about making cool videos for social media. This is about what happens when anyone can create professional quality video content in seconds. Think about what that means for your business. marketing. Think about what that means for everything. I mean, imagine being able to create product demos, explainer videos, social media content, even full commercials just by typing what you want. No cameras, no editing software, no video production teams, just you, your idea, and this AI. And the timing couldn't be better. Video content is absolutely dominating right now. YouTube, Tik Tok, Instagram reels, LinkedIn videos, everyone wants video, but most people think it's too hard or too expensive to make good video content. Well, that just changed forever. Now, let me tell you exactly how this thing works and why it's about to flip the entire content creation industry upside down. Because what Bite Dance built here isn't just another AI tool. It's a complete gamecher. First off, this thing can handle both text to video and image to video. So, you can either describe what you want in text or you can give it a starting image and tell it how to animate that image. Both options work incredibly well. But here's where it gets really interesting. Most AI video generators can only do one shot at a time. Like they'll give you one continuous scene and that's it. Seance 1. 0 can do multi-shot videos. That means it can create videos with multiple scenes, different camera angles, even complex narratives with smooth transitions between shots. Think about what that means. You could literally create a mini movie just by describing the plot. You could make a commercial with multiple scenes showing your product from different angles. You could create educational content that flows from one concept to another seamlessly. And the quality, I'm not kidding when I say this stuff looks professional. The motion is smooth, the physics look real, the lighting is perfect. When I first saw the sample videos, I had to double check that they were actually AI generated. But what really blew my mind is the speed. Remember, we're talking about 41 seconds to generate a 5-second video at 1080p. That's insane. Most AI video tools take 10 to 20 minutes for something like that. Some take hours. This speed advantage isn't just convenient. It's revolutionary because when you can iterate that fast, you can try dozens of different versions until you get exactly what you want. You can experiment, you can refine, you can perfect your content without waiting around all day. And here's something that's going to blow your mind. While other companies are still trying to figure out how to make AI videos that don't look like garbage, Bite Dance solved a problem that nobody else even knew existed. See, most AI video generators have this huge weakness. They can either follow your text instructions really well, or they can make videos that look good and move naturally, but they can't do both at the same time. It's like they have to choose between being smart or being pretty. Bite Dance said, "Forget that. We're doing both. " They built something called a multi-dimensional reward system. I know that sounds fancy, but here's what it actually means. Most AI systems get trained on one thing at a time. Like they learn to make things look good, then separately they learn to follow instructions, make motion look natural. But Seance 1. 0 learns all of these things together at the same time. So it doesn't have to choose between making something look good or following your instructions. It just does both perfectly. And the results speak for themselves. On the artificial analysis leaderboard, which is basically the Olympics for AI video generators, seed dance 1. 0 O isn't just winning, it's destroying the competition. We're talking about beating the second place model by over a 100 points in some categories. But here's what really gets me excited about this. It's not just that Bite Dance made a better AI video generator. It's how they made it better. Because the techniques they used here are going to change everything. They completely reimagined how AI processes video data. Instead of treating video like a bunch of images stuck together, which is what most systems do, they built something that actually understands video as video. It understands motion. It understands how things should move through space over

### [15:00](https://www.youtube.com/watch?v=KAOnu-rHfUw&t=900s) Segment 4 (15:00 - 20:00)

time. This is huge because it means the videos don't have that weird AI jitter that makes you immediately know something was generated by a computer. The motion looks natural because the AI actually understands what natural motion should look like. They also solve the prompt following problem that's been driving everyone crazy. You know how with most AI tools, you ask for something specific and it gives you something sort of close but not quite right? That doesn't happen here. They built what they call a prompt engineering system that basically translates your simple requests into the detailed instructions the AI needs to create exactly what you want. So when you say make a video of a cat jumping, it doesn't just make any random cat jumping. It makes the specific kind of cat jump you had in mind. Um 3 days ago a company called Moonshot AI dropped something called Kimmy Dev72B and it just destroyed every other free coding model out there. I'm talking about beating models that cost hundreds of dollars a month for free. Here's what happened. They took this thing and put it through the hardest coding test in the world. It's called SWE bench verified. This test gives AI models real bugs from real code bases. Not toy problems, real bugs that broke real software. Most AI models fail this test badly, even the expensive ones. But Kimdev 72B got 60. 4% right. That's the highest score any free model has ever gotten. It beat everything else and it's not even close. But here's where it gets crazy. This model doesn't just write code. It actually fixes broken code like a real developer. It finds the exact file that's broken. It writes the perfect fix and it makes sure the fix actually works. I've been testing this for 3 days straight. I threw my worst code at it. Code with bugs I couldn't figure out for weeks. And this thing fixed them in seconds. Not minutes, seconds. The best part, while everyone else is charging you 50 bucks a month for coding help, this is completely free. You can download it right now. Run it on your own computer. No limits, no subscriptions, no nothing. Now, I know what you're thinking. If it's free, it must be garbage, right? Wrong. This thing is built different. The team that made it use something called reinforcement learning. They basically taught it to code by making it fix millions of real bugs. Every time it got a bug fixed, right, it got a reward. Every time it messed up, it got punished. They did this millions of times until it became a coding machine. But here's the part that blew my mind. This model has two brains. One brain called bug fixer that finds and fixes bugs. Another brain called test writer that writes tests to make sure the fixes work. They work together like a tag team. Most coding AIs just spit out code and hope it works. This one actually tests its own work. It's like having a developer and a quality control person built into one model. I ran a test yesterday. I gave it a Python script that was completely broken. multiple bugs, weird edge cases, the kind of stuff that would take me hours to debug. In 30 seconds, it found every bug, wrote perfect fixes, and gave me tests to prove it worked. And the code quality, better than what most humans write. Clean, simple, actually readable, not the messy garbage most AI models produce. The company behind this is called Moonshot AI. They're the same team that made the Kimmy chatbot. But this isn't a chatbot. This is a surgical tool for code. They trained it on 150 billion tokens of real code. GitHub issues, pull requests, the actual conversations developers have when they're fixing bugs. So, it doesn't just know how to code, it knows how to think like a developer. Now, here's why this matters for your business. If you're paying developers right now, this could cut your costs in half. If you're trying to learn to code, this just became your best teacher. If you're building software, this just became your secret weapon. I've already started using it in my agency. Our developers are fixing bugs faster, writing better code, and they're not getting stuck on problems for hours anymore. But it gets better. This thing runs locally. That means your code never leaves your computer. No privacy concerns, no data leaks. Your intellectual property stays yours. Compare that to the big AI companies. They see everything you upload. They train on your code. They could be giving your ideas to your competitors. With KimDev, your code stays private, and the setup is easier than installing a video game. You download it, run one command, and you're coding with the best AI model on the planet. I'm going to show you exactly how to set it up, how to use it, and the specific prompts that get the best results. But first, let me show you what this thing can actually do. Let me give you a real example. Yesterday, I had a client with a broken e-commerce site. The checkout process was failing. Users were adding items to their cart, but couldn't complete purchases, lost sales, angry customers. You know the drill. My developer spent four hours trying to find the bug. Four hours. That's $400 in labor costs, and he still couldn't fix it. So, I fed the code to Kimmy Dev. In less than two minutes, it found the exact problem. A race condition in the payment processing. It wrote the perfect fix, and it even wrote tests to make sure the bug wouldn't come back. 2 minutes versus 4 hours. That's not just impressive, that's business changing. But here's what really got me. The fix wasn't just correct, it was elegant, clean code that made the whole

### [20:00](https://www.youtube.com/watch?v=KAOnu-rHfUw&t=1200s) Segment 5 (20:00 - 25:00)

system better. Most AI models give you messy patches that work but create tech debt. This gave us productionready code and that's just one example. I've tested this on everything. JavaScript bugs, Python crashes, database queries that run forever, API endpoints that return garbage. This thing fixes it all. Now, let me explain how this magic actually works. Cuz once you understand this, you'll see why it's so much better than everything else. Most coding AI models are basically fancy autocomplete. You start typing and they guess what comes next. Sometimes they get it right. Usually, they don't, and when they're wrong, you waste hours fixing their mistakes. Uh, Kimdev works completely different. It has a two-stage process that mimics how real developers actually work. Stage one is called file localization. Instead of randomly changing code, it first figures out exactly which files need to be modified. It reads the bug report. It analyzes the code base. It understands the relationships between different parts of your system. Most AI models just start changing random files and hope something works. Kimmy Dev is surgical. It knows exactly where the problem is before it starts fixing anything. Stage two is code editing. Once it knows which files to change, it writes the actual fix. But here's the brilliant part. It doesn't just fix the immediate problem. It thinks about edge cases. It considers how the fix affects other parts of the system. It writes code that actually belongs in production. And remember those two brains I mentioned, the bug fixer and test rider, they work together during this whole process. Bug fixer finds and fixes the problem. Test writer makes sure the fix actually works and won't break again. This is huge because most coding failures happen when you fix one bug, but create three new ones. With Kimmy Dev, that doesn't happen. The test writer brain is constantly checking the bug fixer brains work. But wait, there's more. This model was trained using something called reinforcement learning in Docker containers. That's fancy talk for teaching it to code in real environments. They didn't just feed it code examples like other models. They gave it actual broken repositories, real bugs that real companies faced. And they only rewarded it when it fixed the bugs completely. Not partially, completely. That means every fix this model suggests has been battle tested. It's learned from millions of real world scenarios. When you give it a bug, it's probably seen something similar before. Compare that to other models that were trained on random code from the internet. Half of that code is probably broken anyway. Kim Dev learned from code that actually works and the results speak for themselves. On that SWE bench verified test I mentioned, here's how it stacks up against the competition. The previous best open- source model got around 40% on this test. Kydeev got 60. 4%. That's a 50% improvement. In AI terms, that's like going from a bicycle to a Ferrari. Even most paid models can't touch this performance. Claude gets around 50%, GPT4 gets maybe 55%. Kim Dev beats them both and it's free. The only models that perform better are the absolute top tier closed source models, Google's Gemini Pro and Anthropics Claude Opus, but those cost hundreds of dollars per month for serious usage. The bottom line is this. These three Chinese AI agents just changed everything. Bite Dance Seed Coder, Kim Dev, and Miniax M1 are giving you capabilities that used to cost thousands of dollars and take weeks to build. And they're doing it for free while American companies charge you hundreds per month. The businesses that figure this out first will dominate their markets. The entrepreneurs who master these tools will build empires. The agencies that implement this will leave their competition in the dust. Don't get left behind while everyone else is still paying for inferior tools. Thanks so much for watching. If you want to get access to all the prompts from today, plus loads of different tests, you can get that inside the AI success lab. This also comes with loads of other trainings as you can see on the lefth hand side here. And inside the classroom, we have loads of different trainings. Additionally, inside the community, we give away new free trainings every single day, like you can see. So, we're sharing new trainings all the time. If you want free gifts daily, this community of 11,500 people and if you want to connect with me personally, feel free to join the AI profit boardroom link in the comments description. Prices are going up at 750 members. So, make sure you lock in your discount right now. Inside there, you can actually request automations. So each week you just post on this thread with the automations you want to see. Then we record a personalized video and create those automations for you. Additionally, you can DM me. You can ask any questions to me. You can see how active this community is. There's just loads of awesome stuff going in there. And also the reason that I've created this is to connect a tribe of serious entrepreneurs interested in scaling their business with AI, making more money, and just winning and learning and growing every single day along the way. And you can see inside this section here, we have all our best agents, templates, workflows, including, for example, all these new workflows that we're adding. Like, for example, a Facebook ad generator, video workflow, perplexity scraper bot, we've got video shorts clippers, AI music generators, one-click book generators, like all

### [25:00](https://www.youtube.com/watch?v=KAOnu-rHfUw&t=1500s) Segment 6 (25:00 - 25:00)

sorts of crazy stuff. And additionally, if you want live help with this, you want live support, if you want to be able to ask us questions and get tutorials in real time, then we actually have three live coaching calls per week. And if you can't watch them, if you can't make them, then you can actually watch back the call recordings here and you get the coaching calls and the Q& A as well. So, feel free to get that link in the comments description. And if you just want to get us to automate all of this stuff for you, feel free to book in a free AI strategy session link in the comments description. What we're going to do on that call is basically look at where you spend your time, figure out what you need to automate, right? We'll come up with a game plan and then if it makes sense for us to do it, we'll just implement that for you and you can become a client of us. So feel free to get that link in the comments description. Appreciate you watching.

---
*Источник: https://ekstraktznaniy.ru/video/6302*