NEW Deepseek Update is INSANE (FREE!) 🤯

Breaking, Deep Sea just released their insane new maths model. We're diving into Deep Seek's brand new release that just dropped, Deep Seek Prover 2. And this isn't your typical sort of chatbot update. This is a mathematical beast that's crushing benchmarks and making other models look average. And it is available completely for free. So today in this video, I'm going to break down what Deep Seek Prover 2 actually is in simple English. The mind-blowing 671B parameter architecture, how it destroys Chat GPT, Claude and other models on maths problems, the free API and chat model you can start using right now, why this matters even if you're not a maths genius, and the fascinating history of DeepSeek and their founder. Now, this model wasn't even announced with a fancy press release, so they just dropped it

on GitHub. Today we're going to be looking at the latest update from Deepseek, which is Deep Seek Prover 2. This is literally just been released and it's basically like a mass model. Now, I'm going to be running you through exactly what it does, why it's important, how it works, etc. And we're just going to get straight into this. Essentially, this is not like Deep Sea Car 2 or anything like that. Just to make it very clear here, what you're looking at is a new sort of different type of model that's more based on logic specifically for maths. Now you can see the benchmarks here. It's also available on hugging face as well. So you can see it's hosted on hugging face and also GitHub as well you can get access to. So it's going to be interesting to take a look at this model. The interesting thing about this as well is that this is a new open-source model, right? So, the thing about Deep Seek is they're constantly releasing new open source models. Now, just give me one second. We'll get back to you on the stream. So, just wait 5 seconds, 10 seconds, and I'll just get back to you in a sec. In the meantime, feel free to post any questions you have. All right. So, if you're wondering about the latest update, basically researcher from Deepseek has said that we just released Deepseek Prover 2. It solves nearly 90% of mini F2F problems. It significantly improves the soda performance on the Putnham bench. It achieves a non-trivial pass rate of Amy 24 and 25 problems in their formal version. Right, this is the

Deepseek release page. So you can see here deepse prover 2 advanced formal mathematical reasoning via reinforcement learning for sub decomposition. All right, I'll be breaking down in a very simplified way what this means in a second. So don't worry about that if you're wondering, okay, what does this mean, etc. What's all this jargon? Don't worry, we're on the same page here. I don't understand half the stuff that was just in that last sentence, but don't worry about that. So we'll come on to that in a second. So you can see here in the introduction on GitHub they said we introduced DeepSeek prover v2 an open- source large language model designed for formal theorem proving in lean 4. Now you can see the benchmarks here. You see the model summary in terms of what it's used for. And if you're wondering, okay, what does all this mean in English? Basically, you can get number one free access via open router to the API. So you can see it was just created on April 30th. So it is a brand new model and it is a 671b parameter model. All right. So if you want to chat with it directly, you can inside the chat here. And then we'll be testing that in a second. And then also you can see what apps it's available with using the API right so you can create an API key for this here and if you want to know okay what apps is it available in so you've got for example root code you got client chat wise all these apps you can use the deepseeek API in and it is free it's a free gift look at that doesn't cost you anything you can switch to standard if you want to pay but wouldn't recommend that just use the free version the one thing that I will say is like when these new APIs get released like for example DeepS quite often what you're going to see is it is slow right it is going to get rinsed essentially but yeah essentially that's how you can use it. Well, let's come on to now what it

means, right? So, let's break it down into basic language and exactly what it means, how to use it, etc. How you can get access and how you know what this why deepseek have really exists as well and also the history of deepse because deepseek have a super interesting history which we'll come on to in a second. All right. Deep sea prover 2 is a mathematical reasoning model. 671b parameters and advanced formal theorem proving. So what is deepseek prover v2? It's a revolutionary AI model designed to transform formal mathematical theorem. So this is not like a chatbot you would design for example for writing content for example. This is like a mixture of experts model. It's fine-tuned for maths. It's recursive problem decomposition. So it utilizes like a sort of a state-of-the-art complex problem solving method. It's also state-of-the-art performance. You can see here how it performs on the mini F2F test benchmarks and how it scores 49 out of 658 challenging problems on the forum theorem proving specifically is specifically designed for like lean forum formal theorem proving makes sense. All right. Now if you want a

simplified version of this, I've got a simplified version for you here. All right. So basically, it's math magic made simple. So what does it mean? It's a super smart maths model designed to solve hard maths problems. So Deep Seek Prover V2 is kind of like a computer brain that's really good at solving maths problems. It can prove why maths answers are right. It was just released. So it was released on April the 30th. And it breaks problems down into like smaller ones, right? So it's very good at splitting hard maths problems, breaking them down, and then figuring how to solve them. Okay. Now, if you're wondering, okay, like who started Deep Seek, what's the history of the company, etc. Let me explain this to you. So, Leon Wfang, he started Deep Seek in 2023. Before that, he ran a sort of money related finance company called Highf Flyer and he just wanted to focus on AR, right? So, in May 2023, Deepseek was really born and they started the company. May 2024, they made Deepseek version two, their first sort of big AR model. That was the one that really came on everyone's radar. And then in January 2025, Deep Seek Car 1 came out and that pretty much blew everything out the water. I think Deep Seek Alpha 2 is going to be even crazier. And then finally, Deep Seek Prover 2 was born and this is focused on maths

problems. Now, in terms of the technical architecture, so how does Deep Seek Prover V2 achieve like all this stuff? So it's got number one 163k token context window recursive surge transformer base coldstar RLE architecture. Basically this is all the technology it's using right so it's really good at like problem decomposition very good at dual reasoning has two different model sizes. So it's available in 7b and 671b and inference optimization. What would be interesting for this is actually if you could test it for coding especially for like hard coding challenges. Coding is very similar to

maps. So that's basically how it works. Now if I had to break that down into a simpler version of how you can use this and what it means etc. And basically Deepseek prover v2 is like having a really smart friend in math so you can break down big problems into smaller easier ones. So stage one is going to be looking at the problem. Stage two is going to be breaking it down. Right? That's the problem decomposition technology that's using. Step three it solves each piece. And step four it puts it all together. So this is really good at math solving hard problems. I think engineering problems stuff like that would be quite interesting as well and you can see how it performs on all the benchmarks right here. So this is proof of a bench which is a comprehensive benchmark for evaluating matical theorem proven capabilities across multiple domains and you can see a breakdown of each right here. So in terms of like educational coverage performance benchmarks comprehensive evaluation and competition grade problems. So, this is really like

achieving state-of-the-art results. The other crazy thing about this is like Deepseek, they didn't release like any crazy white paper or anything like that. They just published this, put it on GitHub, and then released it. And it is absolutely popping off on Twitter. If you check what's trending right now, it's pretty wild. So, you can see some of the examples right here. You know how it's trending. And yeah, now here's how you can use it. All right. So if you want to get started on using this basically you can just grab the API. So for example if we want to start using this we can go into the chat here and then you could give it a riddle or a challenge etc. Obviously is very good at maths. So if we give it let's see if we can find some complex math problems for AI. All right so we'll try this one. I'm not particularly good at maths but we'll take that in. We'll plug that into open routes here. So we're using DeepSeek Provert V2. Again, this is completely free. Interesting that you might be able to connect it to web search as well. I don't think it's going to be particularly useful for that, but essentially then we can start using this reason model for that process. Actually running an error right now, which is interesting. Let's have a look here. There you go. Like I said, the free version usually these new APIs that come out, they break, right? Cuz everyone's like abusing them. But if you switch to the paid version on the standard model, then it will definitely work inside the chat. And you can see the chat here with the breakdown. So you see how it solves the problems, breaks it down in steps, and then starts using its process to figure stuff out. Now, why does this

matter? Number one, it makes learning easier. Number two, a scientist can use this, right? This is not for the average day Joe. And the math helper is more than just fun. It's for solving problems. For example, a scientist can use this to check if their math ideas are actually right, which helps them discover things faster. So it's really good for research, new discoveries, checking ideas. It can help us make better computer programs. So it's really good for coding and that sort of thing. Also just for smarter thinking in general, like anything related to thinking, reasoning, solving problems, everyday puzzles, etc. This model could be really good. Now, if you want to indicate how to get it, so there's a few different ways. You can actually use GitHub as well, and you can open it up in Python, but probably the easiest way to use this is you can just go on to open router. So if you click on open route over here you can get access and then you can see all the breakdown of exactly how it's being used and that sort of thing the recent activity for this and also how it's going. Now you can see here as well the uptime stats. So this is dropping over

time. Let's talk more about what this opens the door for. So key applications the one is academic research. So for example, assisting mathematicians, exploring new theories, automatically generating and validating formal proofs, accelerating mathematical discovery and increasing research productivity. Number two, software verification. So enhancing like formal verification of critical software systems, right? So for example like checking algorithms or thick protocols or mission critical systems. The other option is error detection and correction. So just identifying flaws in mathematical proofs by meticulously checking each logical step helping researchers and students correct subtle errors and then also for educational learning. So it's creating detailed step-by-step explanations alongside the formal proof which basically helps students and mathematicians learn and improve along the way. Now in terms of like everyday applications you could use this for like personal finance that could be an option. Problem solving so like a problem solving assistant. So it could act as basically like a tutor that helps you through solving challenging problems, thinking, reasoning, etc. And educational support. So like it can help students at all levels. Now in terms of the performance benchmarks, you can see them right here broken down. And you can see how this outperforms for mathematical theorem versus GP4, Anthropic, Google Gemini, and SOTA as well. So if we take a look here, if we compare the benchmarks, Deepseek Prover 2 scores 88. 9% on the mini F2F test, whereas GP4 scores 75. 2, Claude 71, Gemini 68, and previous state of the art was 63. So it's a big step in improvement. Now let's talk about what the big differences here. So number one is the context window for the reason model. Number two is the open source part of it. And number three is the architecture for using this as well. And again, if you want to use this with Python, that's a basic way to use it with Python. And then you just install the required packages or you can just get the API alternative and format prompts to start using it. The also the other thing as well I think is looking forward this is only going to accelerate research which help us speed up with developers in AI as well. So that's basically it peeps. That is Deepseek Prover 2 broken down. So if you have any questions, feel free to ask. If you want to get all the notes from today, feel free to get that inside the AI profit boardroom. Every single day there's an update. We just update this with something new. So for example, at Quen 3 just came out. We updated and added a full SAP for that, plus a bunch of useful resources here. You can see as well, Lutra, we've got a full SAP on how to automate stuff with this oneclick AI agent. And this is all the updates just from April. So you've also got updates from, for example, like March, from February, etc. If you go inside the community here as well, you can ask any questions you have inside the community. So if you're stuck with anything, feel free to ask inside the community and we can help you. And additionally, we've got weekly Q& A calls. So if you have any questions, you can just jump on those Q& A, ask any questions you have, and we're happy to help on that, too. And then inside the classroom as well, like we've got so many different courses on AI agents, AI SEO automation, how we've actually implemented AI into our business in a very practical way as you can see right here. And so if you want to learn like exactly how to make money and scale with AI, then feel free to check out this community link in the comments description. It's just focus on making money. And it's a community of members all focused on the same goals as you. Also, if you want to get a free one-to-one SEO strategy session, feel free to get that link in the description. We'll show you how we take websites from zero to 145,000 vis and generate hundreds of thousands of dollars in sales on autopilot. And on this free link building acceleration session, you're going to get everything you need to get results with SEO. will give you a customtailored game plan to grow your business and to get bigger, better results, plus more free leads, customers, and sales directly from