NEW DeepSeek V3.1 Update is INSANE (FREE!) 🤯

Breaking news. We have a new update from Deepseek V3. 1. I'll be showing you exactly how to access it today completely for free. And we're going to get straight into this, my friends. So

basically what's happened here is Deepseek have released this new version called Deep Seek version 3. And essentially what you can do with this is you can use it directly inside the chat for free. This is obviously a Chinese AI that absolutely blew up this year. And I'll be guiding you through exactly how it works. Ashton Lee says, "Hi, Julian. Top G. " So, that's what we're talking about. So, we're going to kick this off with what you can see here, which is V3. 1 base. It literally just got announced a few hours ago. So, you can see the model card collection over here was like a silent release. They've not even announced it on the Twitter. This is breaking news and it's been announced directly on hugging face. You can see the information here. Says Deepseek AI V3. 1 base. Loads of people confused about how to use it and that sort of thing. Other people have been testing it step by step. So, let me show you exactly what it means etc. The interesting thing about this is a lot of

people are actually saying that on SVG Bench which kind of tests all the benchmarks kind of like a G where it gets the API and then runs it through 100 different tests. And you can see here that Deep Seek is actually outperforming many of the other models. One of the best variants of Deepseek so far as well. Now what's interesting as well is when you go on to deepseat now and you go inside the chat and you ask it what version are you number one it replies with which it never used to do before actually if you go on to like previous versions here previous responses you can see that it never used to reply with right and then also directly inside the chat if we go here you can also see it says I'm deepseek version 3 v3 and it knowledge is up to date to July 2024 context window up to 128k tokens that was not there before previously would say I'm deepseek version 3, right? When you have the deep think enabled directly down here. So

very interesting times and we're going to be getting straight into exactly how to use this step by step or we can go inside here inside a new chat on deepseat free to sign up and make sure that you're using chat deepsee and then you can switch between deep think and non deep think and if you use deep think as well it seems to be a lot faster right so we can say here and say okay design a nice ping pong game something like that right and it will start generating the outputs here and we can wait for that to load now in the meantime you might be wondering how to run SVG Gbench. So for example like this deepseek v3. 1. So if you want to use SVG bench then you can run it directly with the instructions from GitHub here. I've got all the notes inside the AR profit boardroom. So all the notes on everything that I'm talking about today is directly inside here. We actually have a full 5hour course on how to use DeepSeek inside the AR profit boardroom too. And also if you have any questions as we go along we are live right now. So feel free to ask. So it's generating the outputs here. And then also important to note on the SVG bench, let me just come back here. So if we go to this section, you can actually see the new benchmarks and you can see V3. 1 is scoring 53. 1% on the list which is actually higher than Gemini 2. 5 Flash, higher than GP5 chat. So it's outperforming many of the top models in the world, even A04 Mini, Flash Light, etc. Kim K2, which is another popular Chinese model. These are all being outperformed by the newest version of Deepseek v3. 1 and this was literally just added a few hours ago. So for example, if we have a look, this was updated 8 hours ago with Deepseek v3. 1. So wait for the chat now, it's actually generating that in Python, which I don't want to do. I say do this in HTML. So I can run it much easier to run. And then if you want to know, okay, what are all the information? specs and stuff like that about Deep Seek? So

release date August 19th 2025 model size 685 parameters context window up to 128k tokens which is still way behind most of the models to be fair like if you look at Gemini's got now 1 million tokens API of claude if you're on level four plan I think it is 1 million tokens as well you can also see the architecture ADA benchmark score cost efficiency so it's 1% better than Opus apparently 68 times cheaper task cost super cheap special tokens availability downloadable now on hugging face under open source licensing apparently and then you can get the rest of the information here. Now you also might be wondering okay what else is there? So it went quietly live. There's been no official announcement yet. So it's been uploaded to hugging face which I showed you a second ago over here. And then also you can see there were announcements in Deepseek's official WeChat chat group which is kind of like an instant messaging group in China that's used quite often. The scale and technical specs. So 685 billion parameters, precision, flexibility, can run the model in multiple formats, hybrid architecture. So this is an interesting one as well. It's like before you had inside Deepseek, you had the thinking mode and you also had the like non-thinking just quick chat mode, right? But these seem to have been built in together. So if we have a look on Twitter for example about Deep Seek 3. 1, you can see here that in the chat, this is a code by Julang took a screenshot. You can see here that basically it supports thinking and search which you previously couldn't do. On ADA it's 71. 6% it's scoring there which is pretty insane. So it's performing really well. This is a brand new post that was submitted 5 hours ago. So it's supposed to be a lot faster and cheaper as well compared to the other models and also it's considered like hybrid. Now the interesting thing about this it's interesting because if you look at chat 5 and the way that's going previously what you'd have to do with the old legacy models if you're using GPT4 is like you have to switch between all these different thinking models right so you use 03 for thinking you'd use something like GPT4 for quick responses etc and you have to switch between them now you have this auto mode that decides how long to think so I can choose okay based on the tasks let's go faster let's go slow etc whereas for example with deepseek you have deep think and non-deink But with this new update, apparently it's kind of auto switching between them as well. Very similar to DPT5. And the other thing I would say here as well is with DeepSeek, I actually I checked on open router. So if you have a look at the APIs, there's been no API updates just yet. So I can't see it on open router. A lot is the same. So you can host Deep Seek locally, but there's been no new update for a local version of Deep Seek V3. 1. So that's not come out. Usually they update their models like every couple of months, but there's been nothing recently as well. Santiago says, "G

seems cool to be OSS. " Yeah, 100%. If you mean like open source, yeah, everyone's going open source. In fact, this is interesting as well. Open source AI is at its peak. This is Victor's post. And you got Quen, you got Google Gemma, Tencent, you also got Mistral, Deepseek 3. 1, etc. So, there's loads of open-source models coming out. And that's basically it for all the new updates. Now the in terms of the token context window so you know how many tokens you can run how long you can have the conversation etc. That was a lot shorter in the previous version. So it's 64k with deepseat v3. Now it's 128k. So it's basically doubled. Then we've got the HTML output from here. Now is it actually any good? That's the question. I haven't tested it enough to really tell you whether it's really good yet at this point or what the big differences are. And that might take a few days to figure out. I wanted to announce the breaking news first so that you are fully aware of it and you get all the information ASAP. Let's test this out. Let's start game. There we go. It's created a pong game. So that's all you need, isn't it? That's all you need. Darren says, "I want more hands-free, fully AI computer autonomous. " So if you want to use Deepseek and its API fully autonomous, hands-free, etc., then what you can actually do inside the air profit boarding, we have this Deep Seek course. All right. Okay, so if you go to the deepseek full course here and we actually have a full SAP on exactly how to use deepseek with browser use web UI, right? So you can use this directly here and with the SAP you can actually get it to control your laptop which is really powerful. You can see here that you can run this locally with Lama and then with Lama you can use browse web UI. We got a full SAP inside the Apuff link in the comments description if you want to check out exactly how to do that. Plus, we show you how to run it locally, but essentially with browser use web UI, you can get that to control your laptop and it is basically what computer use is fully autonomous. So, yeah, the outputs are not bad at all to be fair. And again, it's not using Deep Sea Car1. This is 3. 1. What else we got on the news? I think that's pretty much it to be honest. It is brand new. It's fresh off the press. Let's have a look what people are saying about it. A lot of stuff in Chinese. See what people doing reviews on it. Lots of Chinese stuff. And what's mind-blowing about all of this stuff actually as well is like Deep Seek, they haven't even they don't tweet about it. They don't make a big fuss about it. It's not like they're hyping it up as humans are useless now, etc., which you get a lot from other AI companies in the game. If you look at Deep Seek, like they've not mentioned anything about it. I'm sure they will post a tweet later and it'll be available inside the chat, but like right now they've not even talked about it. There's no information. It's just so chill. So that's pretty crazy as well. and I'm interested to see what the new update comes like. So, thanks so much for watching. That is basically my update on Deep Seek V3. 1. Let's see what else we can try and build here. So, we'll get some posts and that sort of thing. We'll try and build something a bit more stimulating here. So, I'm going to grab this and we'll run a new chat. Wicked Digital Media says, "Jumped into the AR profit boardroom yesterday. " Awesome. Thanks for joining. I appreciate it. The other thing I would say here as well is if we go on to deepseeek API. So you can get API keys deepseeek platform. desec. com and there's no information about the latest information is Deep Seek R1 upgraded, right? There's no information for example owner 3. 1 as well. So this is really breaking information like we got you the update before the company even announced it basically at this point. It's pretty crazy. So it's booting that out right now. And the other cool thing about Deep Seek as well is like you're not going to get charged. There doesn't seem to be a limit inside Deepseek as well. And you can run inside the canvas here. So if you click on run once your response is fully loaded and you can actually preview HTML and anything you build directly here or you can download it and then host it on netify directly too. Breakthrough says open source AI. That's it. And you've got all the files here on Deep See V3. 1 base as well. Also what's interesting is you have an inference provider. So someone usually hosts it and you can test out down here. There's nothing this time around. And there's no announcement on the official hugging face profile for Deep Seek. Like, it's lally just some random person updated a collection. Boom. Let's go. It's wild. So, you can see all these different spaces running different versions of DeepSeek, but you can't see V3. 1 base just yet. You can just see the new model down here. All right. So, that is taking a while to

load. Be interesting as well to tell, okay, what did we get previously versus what do we have now? So for example, if we have a look this previous version, this is dopamine space invaders that we built with deepseek right on the old version. Let's see if we can build it with the new version and just compare the outputs to see which one is the best. So if we go to new chat here, plug in the same prompt and we've got the old version of deepse on the left hand side with the outputs and then we have the new version of deepseek on the right hand side. Now if you actually look at this is interesting as well. So if you look at the thinking time with the same prompt, right? So, we've plugged in the same prompt here and here. You can look at the thinking time. You can see it's actually a lot faster. So, previously it was thinking for 10 seconds. On the new version, Previously, it was thinking for 35 seconds, which is 2. 5 times longer than the new version. So, it seems like the new version is a lot faster. And I do remember that like with Deep Seek, especially R2 and the new releases there, it used to take so long to get a response. Dave Matthews says, "Always the first to bring the best in AI. Thanks, Julian. Happy to help, mate. Thank you for jumping on. Appreciate you watching. What we could also do is compare this versus something like Claude. So, obviously, what would you compare this to? Let's compare it to Claude Opus 4. 1, which is like one of the most powerful coding models in the world, right? And we'll test exactly the same prompt out, and just see which one performs better. I would highly expect the Opus 4. 1 is going to code faster and code better than even the new version of Deep Seek, but let's see what we get here. I do think that Deepseek has got a lot of catching up to do. It's just been so long. And if you look at the context windows and everything else, so we got Claude Opus 4. 1 coding over here and we have DeepSeek V3. 1, the new version coding over here. And then we'll compare it to the old version as well. By the way, whilst we're waiting for this to load, I might as well show you some cool stuff that we've added inside the air profit boarding recently. Over here inside the SCP section, you can see that we're just updating this daily with all the latest new stuff. So, for example, Deep Seek V3. 1, we've got like a whole guide on how to use it. Notebook LM's new updates, the new Google Opal update as well, plus loads of app examples that are built with this and the full SOP. So, if you want to be part of a community where you know you're learning, you're growing, you're getting the latest cutting edge things, then check this community out. It's absolutely awesome. Plus, it's very active. You can see all the new comments inside the community here. You can ask for questions, you can ask anything you want inside here. And then also we run five live coaching calls per week. And inside the class you get access to all my best courses including if you know nothing and you're starting from scratch and you're like maybe I'm a bit too old to learn this stuff or maybe I'm way too behind. Actually if you start there you'll be ahead of 99% of people once you finish. We also have like courses on like how to build AI agents and attend templates agency stuff YouTube stuff or everything you need basically. So you can go as deep or as shallow as you want and you can basically learn AI in one hour a week. is pretty simple and easy. So if we come back to Claude now, you can see here that we have the output from Claude way faster than Deepseek. So Claude is a lot faster for coding and the output looks pretty good. So let's test this out. Maybe not actually. Look at that. It's breaking straight away. So you see how we're just pressing. it's designed the game. So like the enemies land on the player straight away. So you can't even play the game properly. So Deep Seek doesn't have a lot to compete with here. It could be quite easy win if this game just works. Let's test this out. To be honest, this one has failed as well. There's no enemies. There's nothing dropping down. We can't really It's not really playable. That is not working so well as well. So, both of them failed. There we go. I bet you if you I'm very interested now. If we just run this exact same prompt on Gemini, we'll see if Gemini is still the goat. So, we go over here. That's Gemini. Set up the canvas. Plug in the prompt. Exactly the same prompt. And we'll see which one performs better. A journey through existence says are locally hosted models finally going to surpass proprietary models for decoding tools like claw code. I don't expect so the biggest problem is like this cloud code is so powerful isn't it that's the thing it's so powerful and it's I don't see how you can beat unless like the technology gets better but I do think these local models are getting smarter and smarter really the future of it is can you host these models on the phone for example if you look at the way that Google is creating models like Gemma those sort of models are going to run locally on a phone which is going to make your phone and any sort of smaller device which is what most people use these days it's going to make those way more agent and I think That's the future. Whoever cracks that code, which seems to be Google, is probably going to win the race. If you look at Apple, they're struggling so badly with trying to keep up with the AI race. They're not really creating. And there we go. Look at that. All right. Even with sound effects, even though the sound effects were terrible, but you get the point. Like Gemini is still the goat. Claude Code, Claude Opus struggled. Deepseek v3. 1 totally failed and Gemini is still winning. All right, so don't worry about that, peeps. However, if you want something open source, then Deep Seek V3. 1 probably the best. So, thanks so much for watching. If you want to get

access to me personally, connect with me personally, feel free to join the AI profit boardroom. Comes with an amazing community where you can post and ask questions and join 1,200 members. You can also join in the classroom here, get all my best courses, tips, SAPs, etc. We have five coaching calls per week, plus we update this daily with the latest updates, right? So, inside the SAP section, literally as soon as something comes out, we're probably the only community in the world that actually updates with the latest details on exactly what that is, right? Deep Seat comes out, 8 hours later, boom, we got a whole new section on it. So, feel free to join us in here. What else do you get? So, you also get a weekly AI hot seat calls, three live calls per week, all my best AI SOPs, direct access to me, so you can just DM me anytime you want. you get my AI profit playbook. So, basically all the systems I use within my business. And then also, this is on early beta price. So, the price is actually going up after 34 more spots. So, there's only 34 spots left. So, make sure you join now before the price goes up, right? Otherwise, it's going to go up and then you miss out and then you message me saying, "Can you just reduce the price, mate? " And I say, "No. " So, better to sign up right now and hope to see you in there. Now, if you just want us to do everything for you, feel free to book in a AI automation session. This is completely free, but only book it in if you're looking to become a client because on these calls, basically what we do is look at how we can automate your business, how we can save you time, and then how we can implement it for you as a paid service. All right? So, feel free to book that in link in the comments description. And then also, if you want a 21,000 members, all my best free trainings, then you can get that inside the AI money lab link in the comments description. We do post a lot of trainings inside here. The difference is you don't get the coaching, support, you don't get to connect with me personally and you miss out on a lot of the good trainings. So if you want all the best trainings and nothing else, join the AI profit boardroom and I'll see you in there. All

right, Darren says, what else we got here on the questions? Sykes says, is Deep Seek good for coding? You've seen the example today, right? So with Deep Seek, it actually failed on the coding task, but it's just decent for chat in general. However, I would say Gemini 2. 5 Pro is the best model still, right? That outperformed Deep Seek V3. 1. Darren says, "I found Vibe Codin is Yeah, I would agree with that. Yeah, it breaks a lot. " It breaks a lot. If you want some tips on that and the tools that I'm using personally, then just join that profit boarding have updated list to save your time. Richard says, "Thanks. Happy to help, mate. " And Wikipedia says, "Think of it like this. LM are a bit like languages. They all say the same things but slightly different. " I like that example. That's smart. All right, I'll see you in the air profit border in Wicked Digital Media. I know you just joined. Cheers for watching.