OpenAI's o3 and o4 are Here!

03 is right around here, which has a nearly 140 IQ, which is nuts, by the way. And then 01 Pro is 120. 01, you know, the baseline one is like a 90 IQ. And a lot of these are getting to the point where they're at like average IQ or a little bit under average IQ, right? Uh, excuse me. That's crazy. one and two, I cannot believe we have 03 which is in the category of like the top 100ish thousand of people on the planet IQ-wise and then 04 and 04 mini are essentially smart, right? Smarter than the average bear. Hi, my name is Demetri Bonichi and I'm a content creator, agency owner, and AI enthusiast. You're listening to the AI agents podcast brought to you by Jot Form and featuring our very own CEO and founder Idakin Tank. This is the show where artificial intelligence

meets innovation, productivity, and the tools shaping the future of work. Enjoy the show. Chad GPT just released the best models of all time in the thinking side of AI. Let's talk about it. Introducing OpenAI's 03 and 04 mini models. So 03 has been going through many phases. It was an 03 Mini right around when like 01 came out. And honestly, it was pretty good, but at the end of the day, it wasn't nearly what you're about to see. 03 and 04 Mini are the latest in their O series of models that are trained for longer thinking and responses, and they're the smartest models that they've ever released. There's actually something that I saw recently that I wanted to point out. A couple things about this. When we take a look at this from Daria Unumatz, he's somebody who has access to all these models early.

He's like a premier tester and just take a look at the incredible improvement that we've had in IQ tests for these different tools. I mean, it's incredible how much better it's got. And then if we take a deeper look at this chart, just want to open it up for you here. So 03 is right around here, which has a nearly 140 IQ, which is nuts, by the way. And then 01 Pro is 120. 01, you know, the baseline one is like a 90 IQ and a lot of these are getting to the point where they're at like average IQ or a little bit under average IQ, right? Uh, excuse me. That's crazy. One, and two, I cannot believe we have 03, which is in the category of like the top 100ish thousand of people on the planet IQ-wise. And then 04 and 04 mini are

essentially smart, right? Smarter than the average bear. And I'm just baffled at the end of the day as to the improvements. He made a claim that essentially it's going to go from the top 100,000 to the top thousand in the next 2 years. And I think I think there is on to something. It's it's incredible what's occurring now. Some of the use cases we've been seeing here are absolutely nuts. Okay, so just looking at the comparison to its previous models. Oh, we got 01 in comparison to 03 Mini and then the leaps here are there. Right. And a key thing to note about these tools is that what we've seen is 03 with its built-in action calls. By the way, 03 and 04 mini both have action calls and they can actually execute Python during the reasoning chain. So when you're dealing with chat GPT

and you open up 03 right and you ask it a very hard question like can you break down for me the ways in which we can improve the efficiencies in NBA basketball and improve the game from a watchability standpoint. What's happening here is unlike with deep research, it's actually going to be able to do a pretty great job in regards like this. You can see it has 24.90 in regards to the humanities less exam, which is essentially a test that figures out how well somebody would do on an all-encompassing exam on the history of the world, etc. And all we're seeing is a massive improvement for something that isn't even researching for 20 minutes at a time. It's incredible. All of the classic improvements are seen here. You know, we got benchmarks of crazy numbers across the board in comparison to the previous models. It's

higher. Who would have thought? Coding, math, science, function calling, which is very important. And something else that's happening is it can really think with images now. There tools that exist on the planet that can do great with questions like this. You can see it says, I think I missed the ships in the crop. You can see it takes this picture and it says, I took this pick earlier. Can you find the name of the biggest ship you can see? And it actually does a incredible job of this type of thing. I I'm kind of mind blown by it. I'll show you an example in a second. We just finished the 03 mini uh breakdown. And as you can see here, it gave some pretty good examples and showcases of how necessarily to improve the efficiency of the NBA and make it better. And the answers

are correct in my opinion. Keep the ball moving. Keep the basketball players on the floor. I'm sick of not seeing the best players play on the court. Streamline officiating reviews accurate. Give viewers the the feed they want. Accurate. All these are very true. Now, a crazier question would be doing something like this. Find me what church this is and where is it. And what it's going to do is it's actually going to find it based off of the imagery. And not all of these things are publicly available, right? So, what I'm going to be impressed by is its ability to find these things because at the end of the day, we're working in a world and where now we have almost AGI, real augmented. Look at that. It's doing Python. What the heck was that? You can see in the thinking chain, it's running Python.

It's analyzing the data. It's pretty entertaining to me how it's actually going through a real thinking process. And I just want to call this out as it's running. 03 and 04 mini have full access to tools within JGPD as well as your own custom tools via function calling in the API. These models are trained to reason about how to solve problems, choosing when and how to use tools to produce detailed and thoughtful answers and the right output formats quickly, typically under a minute. For example, how will summer energy usage in California compared to last year? The model can search the web for public utility data, write Python code to build a forecast, generate a graph or image, and explain the key factors behind the prediction. This is a huge move forward for things like research, business, science, sports, and visual reasoning. They have examples on

the website that I would dive into that are pretty cool. Cost verse performance 03 mini and 04 mini. You'll see that obviously improvements across the board. Cost versus performance 01 and 03. Just to point out pricing here, 03 is essentially 40 bucks per million output tokens. I'm going to go to the more detailed ones cuz everyone always wants to know that. So, let's go to 03. Okay, so 03 mini $4.40 versus 40. Obviously, this is the higher model. 04 mini is $440. Okay, so it's like the same price as 03 mini, which is interesting. So, what they always do is they seem to like raise the price on the newer one and then the sub optimal new one has a cheaper price than the previous optimal one. Those of you who have access to this will be people like me on the pro plus, pro,

and team plan. And we expect to see the release of 03 Pro in a few weeks with full tools support. I am excited about that. What is that going to be like? By the way, it's going to work trying to figure out what church this is. I really like how deep it's going into this. I actually know off the top of my head what church it is. But this is actually a really difficult question. So that's why I'm testing it with this cuz apparently it's pretty much solved geoging. Like for example, some of the shocking capabilities is it can geog a hike photo. Like for this, it figured out the exact location of this shot and it figured out that it was Porto Cove Provincial Park, east shore of House in British Columbia, which the person in this video did confirm was accurate, which is

crazy. Like, we're geoging now with our AI. I love how it's hyper analyzing different parts of the image. This is actually so cool. You see, zoom center, bottom band. There's a lot of different options like this. Let me pick another one with maybe a better full shot and I can say, where is this picture taken at? Oh, wait. By the way, I think it's actually getting it. I know. Oh, it's looking it up. I can see it found St. Mary of the Angels. That is exactly uh the St. Mary of the Angels uh one right here. So, it is getting close. It's just taking a few minutes. And now I'm showing it a picture of a Independence Day parade. I just love how it's able to analyze these little bits and pieces of the photos. It's actually crazy. St. Mary of the Angels Church in

Chicago's Bucktown. I actually can't believe it figured that out. It took 6 minutes, right? But Geogesser is finished. This is pretty incredible. It figured out what churches was. I I I'm I I don't even know what to say. I mean, I don't know what the practicality of this is right now, but it's actually incredible that it figured this out off of a picture of a ceiling. I I don't even know what to say. So, great work. Honestly, pretty incredible. It even asked if there was another city that it might have been and it could have figured it out. Let's see if it can figure out. This is another Chicago pick that it's analyzing. And I love seeing how it's going through and it's like telling you, okay, therapist bakery. Oh, you know it's the Atropolis. I I I've been to Atropolis before, so I know

it's going to figure it out in a few moments. It's figuring out street signs and stuff like that, which should make it easy, right? Like the the first picture was way harder to be honest. Oh, it's funny. It's trying to I even I mean there's two uh angles here. So, this one I'm surprised it actually took this long cuz it has like an intersection, right? But it's zooming in to make sure it's the right text cuz it's it's like zoom enhance, zoom enhance, zoom enhance, which is funny. Randolph Street. It's really having a hard time with the with the street here, but it does seem to be figuring out it's on Monron Randolph. It's focusing on the patterns of the Greek spots. The Atropolis bakery. H the Greek islandism's right there. Yeah, the Greek islands is right there. So, seems like it is snapped during

the Chicago's annual Greek Independence Day parade in Greek Town. View is facing the intersection of West Randolph Street and North Holstead Street. The banners, blue and white flags, and the round cafe sign on the northwest corner. A familiar Greek Town landmark all line up with the spot. Yeah, it got it. That was incredible. Don't really know what this does for us, but I do know that it showcases the thinking capabilities and the function calls and the actions that it can take. So, if you're looking for the highest level of reasoning, try out 03 now and give it a shot. Thank you so much for watching this episode of the AI Agents Podcast and we'll see you in the next one. Bye.