Deepseek R1 Review & How to Use: The Ultimate Beginner's Guide

Speaker A: In this video we are going to learn about deep seq R1. I personally waited couple of days to gather all the information I could all the data which I can see and make one complete video which will give you the complete picture and full understanding of it. In this video I have focused into what is deep seq R1, why everyone is talking about it, the comparison of deep seq R1 compared to chat, GPT and other top models from a data point of view and from a price point of view. How you can get started using it from a desktop or PC computer. How you can start using it for free from your mobile. And also I'm going to show you how you can get access to API. And in future video I'm going to show you how you can create local uh server of Deep SEQ and start using without Internet. And I'm also going to be showing in future video how you can start using it for coding and SRL for AI agents. Let's get into the video. First important point. This is going to be a very in depth uh explanation video. So I have provided helpful chapter markers in the video description. So highly suggest you in case if you don't want to watch all the video you only interested in specific part of the video. You can click on those chapter markers to navigate into this video. Let's get started. First Deepseek R1 this has been completely revolutionizing currently in last uh one week. You don't need to take my word for it. I just opened up Google Trends report. You can see completely normal but just from January 27th you can see how fast it is growing and it's crazy. For example if I make something like 30 days you can see it's still predicting the graph is going to be going up and up. You can see all the breakout searches like this. Everything is not a one single rising query. Every single thing is a breakout search. So you can see clearly the demand and the popularity for this right now. And if I'm not just uh talking about trends, you go and type deepseek R1. Every single news story is uh, something like this. You can see all the major titles. For example uh, why AI world is freaking out. Sam Altman calls it is impressive and promises to deliver much better work. And Nvidia calls uh, AI excellent advancement. All these things happening and Twitter or X is no nothing different. It's same as that. Every single thing has been going crazy. Everyone is doing and showing all the possibilities with Deep Seq and how better it is right now. Now let me Explain you with a better sense of understanding why everyone is talking about. First deep seek to use it as a chat interface. Almost like um our chat GPT it is completely free. And the cool thing is they are uh open source model which means you can even go beyond using their web chat interface. Download this LLM models a uh smaller distinct version and start using it in your computer or in your software and without you don't need to pay anyone anything. So this is the power of it. For example the main reason LLAMA was getting so much popular was the open source aspect of it. Currently Deep Seq is a open source one and it is rivaling O1 in reasoning. I will be showing the data but this is what everyone is saying. It is as good as the latest chat GPT. That is what everyone is saying. And the cool thing is it's not only for the end user who are just using this product but anyone is using their API or anyone is using this LLM models API. For example a lot of developers will be using APIs to create features uh for the application or if they having some kind of AI based application they will be using this LLM model for their uh reasoning input and everything. The API prices are very very low right now. I will be showing the prices but it is almost a lot of differences between the big models and this model. So that is the main reason it is completely free for everyone right now. You can use it from Internet, uh browser or your mobile app as a mobile app and it has a very low cost API. The main reason they are what they are claiming was I don't want to get too technical into it. They have found a better way to train LLM models in a much more effective way in an obvious endpoint uh view and also in an engineering kind of view. I personally listen to a lot of AI podcast. I listened to AI Daily Brief just today and he was explaining a really good podcast. He was explaining how they are using 8 based uh memory systems, how they are m making multi parameter uh values. All this crazy stuff, technical stuff but just in a simplified way if I have to explain it in one single line. They have found a better way to use already existing computer hardware to get a much better data out of it in a much effective uh prices. Almost like this is how China works in everywhere because Deep SEQ is coming from China. China's main goal has been in many industries always providing a better cheap, not better almost uh working cheaper products. Especially in India and Sri Lanka. That is how we see Chinese products because it Provides the same almost same sort of application but in a comparatively very very cheaper price. So that is what Deep Seq uh doing it round. That is why everyone is talking about it. Let me go through the website, let me close out these uh tabs. You can see search on it. Everyone is going crazy about it. This is the homepage deepseek.com I will leave the link in the description if you want the app claiming that it is live this is the main model I uh will be explaining it is open source and they are directly saying it is rivaling OpenAI's model O1. It is available on web app API. You can download the app by just scanning the QR or going to Google Play store and scanning it or you can start using the chat model directly from by clicking on the start. Now before I go into actually start using the product I want to show you the comparison aspect of it. First I want to compare from a data point of view. Remember these data is provided by them but I'm just not directly taking them because I waited couple of days to see how other big experts, data scientists m uh tell the model like how well they are saying that this works and everything. Then only I am taking this model m and doing a comparison. What I recommend you is you compare it yourself Also if you are already using a pro membership with Chat GPT cloud do the same prompt with both of that. Then you compare which is the better model for you. That is what I would say. I waited because I wanted to see what community react. For example the uh mobile app you can see within a week it almost 1 million download plus and you can see how many reviews they ah are getting. You can see all the things uh happening here. So this is what the power of it. Every single thing is a positive review, almost a positive review. So I'm going to click on seal all review and put it newest. You can see every single thing is completely positive. So this is what users saying. I'm going to take their word but I would recommend to do your own comparison as well. What I did was I'm not a data scientist. You already know I'm a uh just a software basic software engineer with a digital marketing background. So what I did was I just copied every single thing here and went into chat GPT and I pasted the everything and I asked it to help me to understand this table in a simple terms and let me know the most important points. So what that's what I'm going to do. I asked it in chat GPT I Didn't ask it in deep Seq I asked it in chat GPT and it started to break down the table for me. I'm going to skip these things because you can do it. I'm going to go into the most important insights and key takeaway because those are the important ones uh most important insights. Deep seq v3 is the strongest all around model excelling in English Chinese math task as well as in code related chat yes so that is biggest important insight 1 Claude 3.5 performs well in English reasoning, software engineering tax but lags in other areas like Max GPT4O shows strength in English Q and A but less competitive in math and coding uh and Gwen and Llama are strong in specific career but don't match deep six breath uh the depth of it key takeaway. Now let's take a look at it. Deep SEQ is highly versatile and excels across all domain so this is what it the table from the data table because in this data table they never mentioning anything it's just all the values are coming into The Play Cloud 3.5 A strong performance in reasoning and software engineering GPT4O stands out in English QA task smaller models uh perform well in but lack so you can see the key takeaway and important insights from the data point of view almost we could agree that they are uh either getting the top position or they are competing with something like GPT4 and 3.5 I personally feel Claude better for writing and coding GPT4 very good at English and reasoning but Deep SEQ I I huh. Want to make that even better I personally want to see it is better because it is open source we can keep on getting and it is going to be reducing all the AI price very very affordable for us. So those are the important points now let me go into the API pricing because we saw the data comparison now we need to understand the pricing for the API because free user if you are using going to use a chat only interface you don't need to worry about API. You can skip this part go into the next chapter of how to use it. But if you are planning to use the API or if you are a developer you are planning to use the API this section will be important to you because this is the pricing which they are giving right now. Uh for example Deep seek as two models it is one is chat one is a reason a reasoner is the advanced model which reasons any request you give to them it is going to be reasoning then it is going to be providing output so it gives you the context length all these things context length is still bit uh low compared to something like Gemini. And you can see the pricing uh for example for 1 million token cache it means that the data is already available for that. So it is going to be if caching the if if the data is in the cache folder it is going to be charging around 0.014. It is used to be uh at this price they are changing the price model. You can see it is showing the discounted price when the actual price is going to be changed soon. Still the change price is very very low compared to the other model which I will show you in the next screen. But let me quickly show you the other prices as well. So for 1 million token it is 0.$14 and reasonable model. If it's the cache hit it is 0.$14 if it 1 million token cash miss M it it is 0.$55 so output token price is also mentioned here where deep seek chat comes around uh 0.$28 and for deep seq reason it is 2 uh dollar and 2.19 so this is the model pricing. Now let me quickly compare it because that is where we are uh going to be seeing a lot of different results. Okay, let me wait for it. I am going down. I used perplexity for this and you can see a uh deep SEQ chat. I already explained the values we are seeing this. Now let's compare to something like Sorry uh now let's this is the chat we already uh comparison data for example we saw these data like this. Now let's compare to something like Gemini flash flashes. We know that it is a free model for low user and they use give a certain level of API access for free and this is something like um the lowest price. So this is the Flash uh model. But if you are accessing the pro model because Flash model is very limited we already know it's just for basic tasks Almost like an experiment with Google Pro model is one even the pro model is very expensive compared to this for example this is uh 0.5 uh Google Pro comes into $1.25 and token output deep seq reason is 2.919 and this is where it comes to $5 and if I'm taking uh Chat GP into account because that is what most of the people want to see when we are comparing the API cost because everyone who wants a better quality I personally believe it's going to be using either Claude Chat GPT and Google is not far so $15 with ChatGPT 4 where uh. This is uh charging a fractional of that price and for token output this is the amount and for Claude this is the price and this thing I took it from expression comparison with a uh source citation directly asking to it. You can compare it in your own terms also. But just remember the pricing for the API for Deep SEQ has gotten so many developers attention pricing and the open source aspect of it. So you can expect to see many more um your favorite AI apps integrating Deep Seek into their models very soon or they are already doing it. I saw even typing Mind Founder in X uh platform. They really like the Reasoner uh explanation chat and they started to implement within their software aspect of it. So you can expect to see more and more development something like from here as well. Okay, now let me go into how to start using it. In order to start using it there are two ways of doing it. One from a web interface. Sorry, there are three ways of using it. One from a web interface, one from a mobile app one using through API. Let me go one by one. First in order to use the chat model like Chat GPT uh just click on start. Now it will take to this kind of a homepage. This is you can directly comment from your web browser. It is basically chat.deepseek.com and you are very well uh familiar with these kind of interface because there is nothing new Chat GPT, Claude, Gemini Every single interface looks very similar to this. So you can see one too important thing. I have noticed so many people are using this product right now. Everyone is using and there are some. When I started to access this product in the morning when I created an account it even showed a warning message that there are severe DoS attacks on deep seq servers. So you might experience a problem in registering and everything. My recommendation is always use uh email address rather than Google sign in because that is not working for everyone. Use something your personal email that also don't use your private email I mean like um, your name ah company.com use general emails like Gmail, Yandex, Yahoo that tend to work better for me when I was having problems. A small tip like once you come here we have the chat history section Here we have the option to get app profile There is not much about it. Delete all settings A ah very basic settings. You can see I have the control over all the things and basically directly here you can see by default it has selected the R1 model and we can turn on the web search option for to bring relevant data when we want and we can also attach uh, images or docs for example. This is used only for text extraction, almost like a OCR functionality and they are giving the file size limitation on what is the current uh for this. So I have created a prompt here. I wanted to be vague. At the same time I want to give it a complex prompt as well. So I asked it to create a Chrome extension for productivity timer. Let's see how it is handling my request. So I have clicked on chat. Okay, we are back. I wanted to show you the current thing. It worked, but it didn't work at first. I got four errors of operation unsuccessful. Then it took my request. So I am assuming it is because of the demand right now or due to the DoS attacks. But I just wanted to mention that. So when I ask a question and make sure Deep Seek think is enabled, you can see it is going to be showing how the AI is actually thinking. That is the cool part of it. For example I asked this simple thing and it is going into thinking mode and saying okay, the user wants to create a Chrome extension for a productivity timer. Let me break down what they need. First I should outline the basic structure for Chrome extension. It typically includes a manifest file HTML for the pop up CS for styling and JavaScript for functionalities. So it is going into a uh rational thinking format. Then it comes to so the main components would be manifest just uh JSON. This is required for Chrome extension. It defines the metadata permissions and the files involved. POP HTML popup HTML the interface that appears when the user clicks in the extension icon. Pop up CS styles for the pop up to make look nice Pop up JSON so JavaScript and the timer logic user instruction storage so first it under created a whole picture Then it create um went into a deep into the main files of it. Then it is saying explaining the procedure right now starting with the manifest manifest should be version 2 what are the latest so it is going with every single IT identify the overall stuff identify the files. Now inside each files it is saying what it should do for manifest. It's saying all the things it need to do. Now let me read the pop up HTML one single bit uh itself because I'm not going to read everything. It should display the timer start stop button setting for work and break duration. Simpler interface with large number for the timer buttons to control it and input field for settings. Maybe a status message to show if it's work or break time. Everything is coming together. Then it's talk about CSS. Then JavaScript comes into play. I want to read this also because it is going to be the complex part of it and it identified that it also for example more complex. The timer needs to count down switch between work and break periods and persist settings. In using Chrome storage the timer can be managed with set interval updating this display every second. When timers reaches zero, a uh notification should appear and the next phase of work or break should start automatically. So this is also done and it's saying notifications are handled using alarms API in Chrome when time and create alarm triggers a notification service worker uh background JS will handle showing the notification when the alarm goes off and potential issues to consider. Also it is created and you can see making sure the timer doesn't uh drift handling cases, storage, sync testing step all the things. And after that it started to create the actual code itself. You can see it created a structure for me so I can replicate this structure in my local interface. Cool Tip. If you are using Visual Studio you can connect this API and automatically automatically create these files for you. I'll show you that in the next video. So make sure to hit subscribe manifest file. It's giving me the full code of it Pop up HTML it is giving me full code of it Pop up CS it is giving me full coded of it Pop up JS giving me full code background JS everything. And finally it is giving me instruction on how I can go about testing this this uh extension. And it is giving me a uh full feature list of created app uh itself all from one single line of code. And you can see this is a reasoner model. So it thought for 45 seconds then it gave me all the valid output for me. And cool thing you notice is some of uh this section you have the copy button but for some level of coding there is something called run HTML so when you click on it it is going to be giving almost like a display of what it is Currently it is not working because it is not integrating other part of the code into this functionality is just for previewing that HTML. I personally believe they will keep on expanding this to be where we see something like canvas or artifacts from Claude into this also basically we give all these things here we have the copy button regenerate option and also this uh like and dislike almost like chat GPT and a brand new search uh as well. Now let me click on search uh let me get out of the reasoning mode now I wanted to show the search functionalities of this. Hopefully it will work. Let's see if there's errors Also I will show you. So when I uh search for it it is now searching the web also. So not just using their own model, it started to search the web and um, it didn't bring any results for me. Now let me try to continue. Uh hopefully this will work. Everything I wanted to show not only the best of this model but I want to show everything with this because uh it is still not perfect. I mean they are still not able to handle all the demand or they are still struggling with server capacity or something. It didn't work. Let me try something. Maybe they having some issues uh productivity um tips. I'm obviously spelling it wrong to see whether it understand productivity step for work and I want to search this. Hopefully this will work. If not probably I will try it again and see. Hopefully this continues to work. And you can see it went but still it's slow. So you can see it didn't work on the 1st and I'm not sure why that is happening. Let me if I click on here can I see anything? No, it's not working. Then when I do here it is bringing uh the results it showed uh searching options. You can see it is bringing me the results also for example when I click on any of it is it real result? Okay, that is a real result. And when I click on this is it a real result? Okay, this is also a real result. So that is good. For example let me check one more time. This is actually good. So when I asked productivity disk uh it went and gave a in depth answers. I'm not even using deep think. I'm just using the regular model. Then it's giving me a complete answer here as well like I asked tips. It's giving me techniques blocking distraction, decluttering smart goals, practice automative repetitive task. Uh take regular bug for audio listeners. Every single thing which is saying right now it is also giving me how I can what is it sorry how to use it and what are the tools good at it to do this particular stuff. It is just keep on going. It is currently on step number 1515 and it finished all for a single question without the deep seq and bump. Just a search feature. That is how cool this is for me. It is slow. I'm not saying this is 100% fast or you're not going to be experiencing any errors but this is the worst it's going to be. So if it's keep on getting improved you can expect a definite big competitor for Chat GPT and you can also rename it just like Chat GPT for example that is how you rename it and you can come back uh it anytime you want by just simply clicking on it here as well. So that is how you can start using it from desktop environment. Now I quickly want to show you the API section of it. In order to get access to API you can come into the Deep Seq uh website and there is a link for API platform. When you come into the API platform you will see something like here for example you can top up the balance when you click top up you can start minimum of two dollars and you can go up to 500 or custom pricing if you want. I loaded with two dollar you can see then you will see the API key here. I'm not going to show it right now but apart from that we have a detailed usage uh section also here. So it's showing me the price uh monthly usage, deep sea creation and model, how many requests I have made, tokens our money I have consumed. So all the datas are showing here. We can get API key from here to start using with our models. And see this is how um I'm using the API model. You can see I've already connected to versus code. This is the next video it is going to be so in this video I have explained everything you need to get started off it. Next videos in this series I'm going to be focusing on how to install Deep Seq R1 locally in your computer how to use Deep SEQ for coding with very very low affordable prices from Visual Studio and other coding platforms. Also something like Bold D A Y I How to use the next video how to use deep seq for open source AI agents something alternative to search uh open AIs, chat GPTs operators so these videos is going to be the this field again has changed a lot. We are just entering into 20. Do let me know what your thoughts about deep seq R1 and how you seeing everything evolves around and what you are using going to use chat GPT or cloud or Deep seq. Let me know in the comments and if you enjoyed this video. Also do leave a like and if you want to watch the next set of video subscribe and get notified. Thank you.