This NEW Chinese AI DESTROYS GPT-o1 & DeepSeek? 🤯

China just dropped the biggest AI bomb since Deep Seek. Yesterday, BYU open sourced their Ernie 4. 5 model with 424 billion parameters. This thing beats GPT40 in almost every test. And get this, it's completely free and open source. The AI world just got turned upside down. We need to talk about this right now. 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. So, what exactly happened? Yesterday, June 30th, 2025, BYU just open sourced their ENIE 4. 5 model family. And this isn't just another AI release. This is massive. We're talking about a model that has 424 billion total parameters with 47 billion active parameters. That's huge. But here's where it gets crazy. Looking at the official benchmarks from BYU, this model is beating OpenAI's best models in key areas. Let me break down the real numbers for you. In Math Vista, which tests math reasoning with visual context, Ernie 4. 5VL 424B A4B scored 78. 9. Open AIO1 only scored 71. 4. That's a 7. 5 point win for the Chinese model. in docu which tests how well AI can answer questions about documents. Ernie 4. 5 VL424B A47B scored 93. 1. Open AIO1 scored 81. 0. That's a massive 12point difference in CCBench for visual knowledge. Ernie 4. 5 scored 83. 9 compared to Open AIO1's 79. 7. Another solid win. And here's what's really impressive in document and chart understanding. Ernie 4. 5 consistently outperforms chart QA 86. 8 versus open AI1's AT4 OCR bench 872 verse 761. These aren't small gaps. Now here's what really blew my mind. Looking at the texton performance Ernie 4. 5 300b47B is competing directly with GPT 4. 1 Deepseek V3 and other top models. In many math benchmarks it's winning GSMK 96. 6 CMATH 96. 7 math 500 96. 4. These are incredible scores. In reasoning tasks, it's holding its own too. BBH 94. 3 drop 91. 1. These numbers put it in the same league as the best models available. You're getting worldclass performance for a fraction of the cost and now it's completely open source. But let's talk about Deep Seek 2 because everyone's been talking about Deep Seek V3 lately. Well, looking at the official benchmarks, Ernie 4. 5 is competitive or beating Deepseek V3 in many areas. In reasoning tasks like BBH, Ernie scored 94. 3 versus Deep Seeks 91. 0. In math tasks like CMATH, Ernie scored 96. 7 versus Deepseeks 94. 8. The competition between Chinese AI models is fierce and that's pushing performance higher while driving costs down. This is huge because it means anyone can download and use these models. Developers can build on top of them. Companies can integrate them without paying massive API fees. Students can learn from them. The barriers to entry just got demolished. Now, let me tell you what makes Ernie 4. 5 special. Technically, it uses a heterogeneous mixture of experts architecture. The largest model has 424 billion total parameters, but only activates 47 billion for each task. This makes it incredibly efficient while maintaining massive capability. It's also truly multimodal from the ground up. The VL models can understand and process text, images, audio, and video all in one model. You don't need separate models for different types of content. And here's something cool. The post-trained models have both thinking and non-thinking modes. In thinking mode, you can see the AI's reasoning process step by step, just like the reasoning models from OpenAI and DeepSeek. The training approach is also innovative. They used stage training, first building strong text capabilities, then adding multimodal features without compromising the text performance. Most companies struggle with this balance. BU figured it out. Now, let's talk about what this means for the AI industry. We're seeing a pattern here. Chinese companies are releasing powerful open-source models that compete with or beat the closed source American models. And they're doing it at a fraction of the cost. First, it was Deep Seek with their R1 model. Now, it's BYU with Ernie 4. 5. This is putting massive pressure on OpenAI, Google, and Anthropic. They can't charge premium prices when you can get similar or better performance for almost free. And here's what's really interesting. BU's CEO, Robin Lee, used to be against open source. He believed in closed models, but Deepseek changed his mind. He said, "If you open things up, a lot of people will be curious enough to try it. This will help spread the technology much faster. " That's a huge shift in thinking from one of China's biggest tech companies, and it shows how much the landscape has changed. Now, I know some of you are wondering about the limitations, too. Ernie 4. 5 isn't perfect in every area. In some coding benchmarks like Live Codebench, it scores 38. 8 compared to other models that might score higher.

It's also optimized for Chinese language tasks. So performance might vary depending on your specific use case. But for multimodal tasks, document understanding, math reasoning, and visual perception, this model is incredibly strong. And the fact that it's free and open source makes it a gamecher for most applications. Here's what you can do with Ernie 4. 5 right now based on its benchmark performance. Document analysis is where this model really excels. Those doc VQA and chart QA scores show it can handle complex business documents, financial reports, and technical charts better than most models. The math and visual reasoning capabilities are also incredible. That Math Vista score of 78. 8 means it can solve complex math problems that involve graphs, charts, and visual elements. This isn't just textbased math. This is true multimodal reasoning. Video understanding is another strength. The MV bench and video MME scores show it can analyze video content, understand temporal sequences, and reason about what's happening across time. That's huge for content analysis and video processing applications. You can download these models right now from HuggingFace. The main models to look for are Ernie 4. 5 VL424B A47B for the full multimodal experience or Ernie 4. 5300B A47B for texton tasks. You can run them locally if you have enough compute power. You can use them through BYU's API or you can try them through Ernie bot which is BU's chat GPT equivalent. The model family is massive. There are 10 different variants. You've got text only models from 0. 3B all the way up to 300B parameters with 47B active. Then you've got the multimodal VL models at 28B and 424B parameters. Some have thinking modes, some don't. Some are base models, some are post-trained. You can pick exactly what fits your needs and hardware. And here's something exciting. BYU is already working on Ernie 5. That's expected to launch in the second half of 2025. If Ernie 4. 5 is this good, imagine what Ernie 5 will be like. But here's what I think is the biggest takeaway. The AI race is no longer just about American companies. Chinese companies are not just catching up. They're leading in some areas. And they're doing it with open-source models. This changes everything. It means access to powerful AI is no longer limited to big tech companies with massive budgets. Small startups can compete. Individual developers can build amazing things. Students can experiment and learn. The democratization of AI is happening right now. And Yi 4. 5 is a huge step in that direction. Now I know some of you are wondering about the geopolitical implications. Yes, this is a Chinese model from a Chinese company and yes, there are ongoing tensions between the US and China in tech. But the model is open source. The code is available. You can see exactly what it does. And here's the thing. Good AI benefits everyone. Whether it comes from America, China, or anywhere else. Competition drives innovation and consumers win when there are multiple strong options. What I'm seeing is a new era of AI development. One where open-source models are competing directly with closed source ones. Where smaller companies can challenge the giants, where innovation is happening faster than ever. and Ernie 4. 5 might just be the model that proves open source AI can not only compete with proprietary models but beat them. The question now is how will open AI, Google and other American companies respond? Will they double down on closed models? Will they embrace open source? Will they have to significantly improve their models to stay competitive? Whatever they do, they can't ignore what just happened. A Chinese company just released a model family that competes with or beats the best proprietary models in many tasks for free. That's a game changer. Here's what I recommend you do right now. Go try Ernie 4. 5. Download it if you can. Test it on your use cases. See how it compares to what you're currently using because this might be the model that changes your AI workflow forever. The AI world just got a lot more interesting and a lot more competitive. Ernie 4. 5 isn't just another model release. It's a statement. It's China saying they're not just players in the AI game. They're leaders. And if you're not paying attention to what's happening in Chinese AI, you're missing the biggest story in tech right now. Now, let me tell you about some specific use cases where Ernie 4. 5 really shines because knowing the benchmarks is one thing, but understanding how to actually use this model is what matters. Remember that DOC VQA score of 91? That means it can read complex documents, charts, and graphs and answer detailed questions about them. If you're in law, finance, research, or any field that deals with lots of documents, this could be huge for you. That Math Vista score of 69 means it can solve math problems that involve charts, graphs, and visual elements. Think about applications in education, engineering, data analysis. This is visual math reasoning. Um, the multimodal capabilities mean you can

feed it text, images, audio, and video all at once. Imagine analyzing a business presentation that has slides, voice over, and video clips. Ernie 4. 5 can process all of that together and give you insights that single modal models can't. The thinking mode is also fascinating. Just like GPT01 and Deep Seek R1, you can see the model's reasoning process. This is huge for complex problem solving where you need to understand how the AI reached its conclusion. But here's what I find most exciting about the open source release. Developers can now build custom applications on top of this model. They can fine-tune it for specific industries or use cases. They can combine it with other tools and services. The possibilities are endless. And because it's open source, you're not locked into any single provider. You can run it on your own infrastructure. You can modify it as needed. You have complete control over your AI stack. This is also massive for developing countries and smaller organizations. They now have access to worldclass AI without paying premium prices. That levels the playing field in a way we've never seen before. The timing of this release is also perfect. We're in the middle of an AI arms race. Companies are spending billions on AI development and deployment. Ernie 4. 5 just made a lot of that spending unnecessary. Why pay thousands of dollars per month for GPT4. 5 API access when you can get similar performance for free? Why lock yourself into a proprietary system when you can use an open alternative? This puts enormous pressure on the big AI companies. They'll have to either significantly improve their models or significantly drop their prices. Either way, consumers win. But let's also talk about the technical innovation here. The heterogeneous architecture is genuinely impressive. Most companies struggle with multimodal training. Either the text performance suffers or the visual performance suffers. BYU figured out how to do both well. The efficiency gains are also remarkable. 47% model utilization during training is industryleading. That means less waste, lower costs, and better performance per parameter. And the stage training approach is smart. They train the text capabilities first, then added multimodal features. This ensures a strong foundation before adding complexity. These aren't just incremental improvements. These are significant technical advances that the entire AI community can learn from. Ernie 4. 5 isn't just a new AI model. It's a paradigm shift. And if you're not prepared for what comes next, you're going to be left behind. The AI revolution is accelerating and it's happening in ways that nobody predicted. Are you ready? Julian Goldie reads every comment, so make sure you comment below and let me know what you think about Ernie 4. 5. Are you going to try it? How do you think this changes the AI landscape? If you want to scale your business and get more customers while saving hundreds of hours with AI automation, you need to check out my AI profit boardroom. This is where I share the latest AI strategies that are actually working for real businesses. And if you need help with your SEO strategy, I'm offering free SEO strategy sessions. The link is in the comments and description. We'll analyze your website and show you exactly how to get more organic traffic and leads. Plus, if you want step-by-step tutorials and processes for using AI in your business, check out the AI Success Lab. We have over a 100 use cases, detailed SOPs, and a community of 13,000 members who are all scaling their businesses with AI. Thanks for watching, and I'll see you in the next video where we'll dive even deeper into the AI revolution that's happening right now.