ChatGPT 5.2 is INCREDIBLE. Here’s how to use it in 14 minutes

Chad GPT 5. 2 just launched and I am absolutely blown away in almost every measurable category except for a couple I'll go over. It is a massive step up. In this video, I'll show you everything that's changed, show you how to get the most out of this model so your workflow dramatically improves, and even give you a couple prompts you can start using right now that will give you incredible results. This is the AI model you need to be using if you want superpowers. Let's get into it. So it just dropped

and of course it dropped while I was in the middle of a flight to Mexico. So that's why I have a different background here. But here are the benchmarks. It has absolutely blown away the benchmarks in almost every eval. It is a big step up over 5. 1 on knowledge work on software engineering which we'll do a deep dive into on scientific questions on tool use on a whole bunch of things. Now I've said it many times before you need to develop your own benchmarks. I honestly believe a lot of these benchmarks they use are total BS. I think it's gamed by a lot of these companies, but we'll go into how well it actually performs on a lot of true tests. Now, here is what you need to

know. Here are the strengths and weaknesses of 5. 2. One is more humanlike ideas. My two main use cases with AI are coding and business partner coming up with creative novel ideas and 5. 2 is the best model ever created when it comes to being a business partner and coming up with novel creative ideas. This truly feels like a creative partner. So, as I'm planning my many businesses out, whether it's my SAS, whether it's my community, whether it's other products I'm building, I run it all by my AI model. And 5. 2 has come up with the best, most interesting, most humanlike ideas I've heard from any model so far. Let me give you an example of this that you can take right now. So, here's a

prompt. I'll put it down below. I really highly encourage everyone watching this video take this prompt and run it for yourself. But this is a prompt I use to test the ideiation and creativity of the model. This helps me generate business ideas. I like to build micro SASS. I like to start little tiny micro companies every single week. And so I'm constantly looking for new ideas to build out. This prompt is the one that helps me do that. You are my brutally practical co-founder. I'll give you my background and interests and it's your job to generate an app and business ideas that are actually shippable and sellable, not generic AI habit tracker trash. So, we're giving it a role. We're giving it context about ourselves. So, I give it all my skills, my career, things about my interest and obsessions, my reach. So, who's my audience, who I'd be selling to, things I wouldn't want to build. So, giving it some sort of constraints, and then giving it rules for what I'm looking for an output, ideas, a name, a pitch, things like that. You should take this right now and fill it out with all your information and give it a test cuz this will help you generate amazing ideas. But here's the thing. The output I got from this prompt from 5. 2 thinking, which the thinking model of chat GBT is always my go-to. I think it's like the best model there is. The output was actually the strongest I've ever got from these models. The previous best was 5. 1. I even preferred 5. 1 thinking to Opus when it came to business planning. But here's the issue I have with most AIs when it comes to this prompt is they give me slop ideas. They give me ideas for businesses that kind of sound good on paper but are not actually ideas anyone would use. Right? Here's a few ideas. It gave me a prompt regression test. So an app that actually tests my system prompts in my apps as the product evolves to make sure the system prompts are still good. That's a very good use case. That's an issue I ran into with my app creator buddy. Vibe Code Auditor where you give it your entire repo and it helps you do a safety checklist. That's a big issue a lot of people have is around security. So this is basically a security auditor for your vibe coded apps. That's something people actually need. These are all really strong ideas. But now let's compare it to some other models out there where I've been getting kind of slop ideas including 5. 1. Right? So this is just one model earlier. Let's look at the ideas. Creator Deal Desk. An AI deal desk that reviews creator brand contracts, flags, traps, suggest better terms. That like sounds like a good idea, but the audience for that and people needing that is not very high. Let's go to the next one. Vibe build roulette. One tap weekend build brief generator for vibe coders who don't know what to ship next. Like these sound good, but no one's going to want a vibe build roulette. It's kind of AI slop where it looks and sounds kind of good, but when you think about it deeply, it's not really that good. It's like when AI image generation came out and the picture put out looked pretty good. Then you'd look and they'd have like seven fingers and eight toes. This is like a toes uh response right here. Now, let's even compare it to Opus. So, I gave the exact same prompt to Opus 45, which many believe is the strongest model before this. And it's kind of the same thing. kind of slop ideas, vibe check, screenshot your code, get instant, is this good enough to ship feedback. That's not anyone's true workflow where they're screenshotting code and giving it to AI. That's not a real idea. That's kind of a slop idea that sounds good, but no one would actually use it. Tinder for app ideas, right? Swipe on concept, see what the vibe coder community actually wants to build. again sounds kind of and that one doesn't even sound nice to be honest with you, but it's slop at its core. It's not an idea anyone would actually want to use or build out. The ideas I get from 52 actually feel like they come from a human being. It feels like the main source of these ideas came from like a human being's brain. So, this

instantly after my test became my go-to daily driver business partner, which other than coding is my number one use case for AI. You should be doing the same thing, too. The best exercise I ever did with AI is every knowledge task I do on a day-to-day basis, I feed it through AI to see how it can improve those ideas. This is truly the first model where the ideas it's giving back to me and the ways it can improve my knowledge workflows actually feel amazing and humanlike. So this is really big. In a lot of instances it's much faster. So for instance that prompt I just showed you it did it in like a couple seconds as compared to 5. 1 thinking. So I was really impressed with the speed on some of the prompts. It seems to know when to think faster. Also it has incredible tool use. It is able to string together many different tool uses into one prompt. Let me show you an

example. So, I'm going to give you this prompt down below as well. I think this is a prompt everyone should be using, especially if you're an investor and I think investing is super important in the age of AI. Basically, I made it my AI equity researcher where I gave it a bunch of AI stocks and I wanted to do research on those stocks. I wanted to give me all the information on those companies. I wanted to give me market data. I wanted to remove fluff. I wanted to create spreadsheets on all the tickers. I wanted to give me bull cases, bear cases, 100 different things that are going to require many different tool uses. And the results are stunning. The results are amazing. I mean, no other AI model, 5. 1, Opus, no other ones, gave me results nearly remotely as close as this. So, let's look over on the right to see the reasoning it did. So, it worked for 26 minutes 48 seconds. And what it did, and you can look, I highly recommend looking at all the thinking of all the prompts you give it cuz it's really, really stunning. It went around. It checked many different websites for all the different tickers and all the information about those tickers. You can see every step of the way it's checking like 14 or 15 different websites to gather data. It's looking for earning results from different websites. It's looking for operating income and financial reports. It's getting stock charts and stock data. It's searching for historical data. And then it's downloading CSVs from these different websites. Look at this. Clicking download file. So, it's doing research. It's downloading files. I assume they're downloading files in the OpenAI servers. Maybe that's why they have to raise like $98 trillion to run their company. But they're downloading these files. They're pulling the data from the files. And then you can see here after the course of 26 minutes, it told me everything it did. It created a bunch of spreadsheets for me. So we can actually open this up. Let's open this up right here. And you can see a lot of these research spreadsheets it made me. So you see all the companies I listed. It's every company I listed is right there. the company, the sector, the current price, the performance over many different periods, the PE, and then it even did research for me. Look how amazing this is. Like, I'm actually truly blown away by what it did here. The moes for each one, the risks. It gave me three different bull cases for each one of these companies, three different bear cases, and what would change its mind? That is actually incredible. I'll give you this prompt down below so you can run this yourself and actually test how amazing this tool use is. But from a tool use perspective, it is incredible. You need to be giving it tasks that makes it do a lot of different things at once cuz it's knowledge around tool use is beyond anything. It's like Opus 45 for coding. This is that same leap. On

top of that has the best memory system in AI, which has been the case with every OpenAI model for many models now. When talking to Chat GBT, it just remembers the most details about me more than any other model I talked to. It really is incredible. This really probably is Chad GBT's biggest moat by far is the memory. It just remembers the most minute details from all our conversations. And on top of that, it's very humanlike and fun to talk to. I think that's something that Chad GBT lost when it went to 01 and 03 is it lost kind of some of the warmness and personality of 40, which I think they have finally successfully brought back for 5. 2. The personality is the strongest in any of the models I talked to. So, those are the strengths. What

about the weaknesses? It can't be all rainbows and sunshine. Well, it's true. My number one use case for AI is coding. I code 12 to 14 hours a day, every single day using AI, basically exclusively. Opus 45 felt like the biggest leap in AI coding of my entire lifetime. It was unbelievable. 5. 2 is very, very strong, but it unfortunately is not Opus 45. In fact, I would go as far to say is it's actually smarter than Opus 45, but I still don't prefer it for coding. The reason being is it's just lazy. It just feels lazy when I use it. When I use Opus and give it a command, it not only does exactly what I want it to do, it does more. It thinks of more ideas and actually, hey, by the way, I built it out for you, but here are other things I kind of want to build out as well. Is that cool? GPT 5. 2, on the other hand, my most given prompt to 5. 2 2 as well as 5. 1 and five codecs was okay well then do it. I'm constantly nudging chat GPT to finish the task I give it when it comes to coding. Constantly I'll say do this and they'll go okay I'm thinking about doing this and I go okay just do it then. What are you waiting for? Do it. I'm not even in plan mode. Just do it. It feels lazy. It feels like that senior dev who's just very smart but doesn't want to go the extra mile for you. When I managed a team at a big tech company, I preferred the employee that who would maybe wasn't as smart, but went the extra mile and worked at the edge of authority, right? They did a whole bunch of things and they asked for forgiveness rather than permission. I had smarter employees than that star employee, but I didn't like them quite as much cuz they didn't go that extra mile and they constantly needed handholding. That's exactly how I feel about 5. 2 where I'm constantly nudging it. I'm constantly handholding it. It feels smarter. It finds errors Opus can't even find. But I just have to give it so much more attention to make sure it completes the task. Until Open AI fixes that issue, I'll have to stick with Opus for coding purposes. Maybe it's cuz we didn't get the codeex update for the model for 52, but I mean at the same time 5. 1 Codeex had the same exact issue. From an artifact perspective, the artifacts in Chad GBT are not quite as good. And what I mean by that is when you build apps inside like actually like claw. ai, AI. It actually shows you the apps and shows you a whole bunch of cool things inside the window. Chat GPT has the canvas and a couple other artifact functionality, but it's still not quite the same where I feel like it's not doing as many things proactively inside the chat window for me. When it comes to artifacts, they also bumped up the price. Chat GPT 5. 2

is more expensive now. And you can see it here. They bumped it up by a good amount, almost 50% more expensive than 5. 1. So, that's unfortunate to see. I'll have to keep 5. 1 through the API in a lot of my apps just to keep prices down. So, let's talk about the use cases. Here

is my updated use case list for all of AI. I put in 5. 2 into a lot of these categories cuz it is now the strongest one for a lot of these. So, coding, I'm sticking with Opus 45. Straight up answers to questions. I took out Gemini 3. I'm sticking with 5. 2 Instant. I think 5. 2 Instant is a fantastic day-to-day daily driver model for just asking simple questions. current events because of Gemini being plugged into Google seems to be much better getting current up-to-date information creative writing business planning as I showed you 5. 2 thinking is by far best in the business biggest leap so far you need to be using 5. 2 thinking for the prompt I showed you earlier and for all your business and creative thinking media I'm sticking with Google for most of these social media stuff Gro is best cuz it's plugged into X and then 5. 2 Pro I'll have a dedicated video for 5. 2 2 Pro. So, make sure to subscribe and turn on notifications for that. But that is now the best hyper premium model there is. And then deep research because of that equity use case I showed you where I was creating these incredible spreadsheets and doing all this other research. I have to go with 5. 2 thinking for the deep research. This was I was blown away by this AI equity use case. You have to steal that prompt from down below. 5. 2 thinking is a massive leap. It is brand new. It just dropped. You have to be using the newest tech. the moment it drops, you have to. There's so much opportunity with things like I showed you here, with building, with creating reports, with finding information. There's so much opportunity when these new models drop, and you need to be using them. If you learned anything at all, make sure to leave a like. Make sure to subscribe. All I do is make amazing videos about AI. Turn on notifications so you get them the moment they drop. I really appreciate you watching. It means the world you'd watch my videos. and I'll see you in the next