ChatGPT o3 Mini - Best Model In The World & It's FREE

All right, so OpenAI just released O3 mini, the smartest, the best model that we have ever seen according to all the benchmarks. In this video, I'm going to show you how to use it, which we're just gonna do right here. If you're in a paid plan, you simply go into here and you can switch to O3 mini. But as you can see, this is not exactly intuitive. Matter of fact, I would argue it's quite the opposite. You have O1, you have O1 Pro, you have O3 Mini, you have O3 Mini High, then you have Deepseek, which essentially does the same thing with what should you be using and what decisions you make here, depending on your budget. So in this video, I want to answer two questions. Well, kind of three questions. The first one, which one should you be using if you're on a free plan? If you're not willing to spend a single dollar, what is the smartest AI model that you could be using today? Second question, what should you be using if you're on an unlimited budget and you just don't care what you spend? Give me the best model in the world. I want that. And then thirdly, what do you spend? Or like, which model do you pick if you're on a $20 budget and you're on a plus plan or similar? We're going to answer all of these and we're going to do it based on benchmarks. This video is not going to be about running some random test cases or doing one off prompts or something like that. We do that in separate content here. We're simply going to look at all the benchmarks that came with the release. Because the four focus of these thinking models, as you might know, is really more science driven, more mathematical and coding focused use cases. So those are really hard to judge one by one. We're going to rely upon the benchmarks here. Vibes are a different story. Okay, so data driven recommendations on what to use. What is the best model in the world today for you to use? Let's go. All right, very simple. I prepared an overview for you. Okay. With pen and paper here, I compiled all the benchmarks. I'm going to show them to you in about 15 seconds. I just want to spend 10 seconds complaining by how messy this release is. If you try to look into this yourself, it's a mess. Like seriously, not just the naming, I'm not just talking about the fact that we have 01, 03 mini whatever. 03 mini high. Sure. Like, I'll cut them some Slack for that because, you know, like, Deepseek only has one model and they named it R1, so that's simpler. But the benchmarks that they released here. Jesus. Like, the ChatGPT Pro blog post doesn't even include all the benchmarks that are included with the O3 mini blog post. So that's kind of messy. But then also, these code forces benchmarks, they're just. Here they're expressed in ELO values, and here they're expressed in percentiles in the world. So then I had to go in and I had to, like, look up what each ELO means in terms of percentile, and I had to translate them so we could compare apples to oranges. Apples to apples, because this is apples to oranges. Long story short, here is everything that I came up with for you in one sheet. And let me make my recommendations based on this before I do this. Quick explanation of what you can see over here on the screen. Basically, you can see all the models that matter the most right now. You can see 01, ChatGPT's old reasoning model. Then you can see 01 Pro, the one that is only available to the 200 Pro plan. Then you see O3 mini on the low setting, the one that they just released. And you have O3 mini medium. By the way, low, medium and high refers to how much reasoning the model is doing. Okay, so if it spends more time reasoning, more time thinking about something, it just performs better across all the benchmarks and does better. But as you can see, it also takes place more time. Matter of fact, I couldn't even test O3 mini low in ChatGPT because it's not available. It's. That's. That's an API thing, an O3 mini load. Their blog post says it's not available. What you get on the free plan is O3 mini medium. Okay, so we have O3 mini medium, then we have O3 mini high. And then we have deepseek R1, which, you know, the entire world is talking about now and a lot of people are using, and we're comparing all of the benchmarks that have been published here to each other in yellow with this wonderful Sharpie marker that I got from California last year. I circled the highest values on here. Okay, so as you can see, O3 mini high across the board. Really amazing. Okay, that's the one we want to be looking at. And at the bottom, this speed number, this is a purely subjective benchmark that I created here over the past hour. Okay, so take it with, like, not just A grain of salt, but a truckload of salt. But I think it's worth something. I basically went ahead and I ran the same prompt here through 01Pro three times, through all three mini free times, all of these three times. And I averaged it out and these are the results I got in seconds. Okay, so as you can see, the old 01 was the fastest, and 01 Pro, as expected, is by far the slowest here. Okay, so those are the results. Now let's get to the recommendations based on these results, based on the data, what should you be using if you have an unlimited budget, if you don't care what you spend, you just want the best model out there? Well, you would probably be on a 01 on a ChatGPT Pro plan, but there's no real point in doing that anymore because as you can see, O3 mini on the high setting actually outperforms ChatGPT Pro on these benchmarks. Now, there's one caveat to this which I do want to highlight, and that's the fact that the blog post that was published with ChatGPT Pro, these numbers on, for example, like O1, they don't line up with the new O3 blog post. That's why I'm saying this one. This is a little confusing. And the O1 Pro evaluation, I'll caveat this because look at that. Competition math, 78 for 01. And then if I go to the O3 mini competition math, same benchmark, right? All of a sudden 01 has 83 again. Over here it was 78 and now it's 83. I don't know what's up with that. But basically I took these numbers for O1 Pro. These are the only benchmarks I could find on the Internet and from OpenAI. And if I compare them, 03 mini on the high setting just wins across the board. So there's no real point to pay for the $200 plan if the only thing you care about is the smartest model out there. There's a lot of other points, you know, no rate limits, for example, because O3 mini Pro is rate limited. I'll put it in the top comment in the description. They keep changing this. I'll pin it in the top comment. That's what I'm trying to say. And the other models underperform it. So as you can see, no point for the $200 plan. If you want, operator, if you want, no rate limits. If you want unlimited, advanced voice mode, unlimited sora, all these wonderful things that may be well worth it. I think especially Operator is for many people, for the smartest model, you don't need it. Okay, so then let's move on over to the second question of this video. Which model should you be using if you Want to spend $20 on a premium plan or if you are spending $20 on a premium plan? Well, as you can see, O3 mini on the high setting has the highest performance, so that's the one you want to be using. You get better benchmarking results across the board then deep seq R1 and you get a better speed. Right? This is in seconds, so lower is better. So over here you can see in this row, I like this. 60 seconds is faster than 95 seconds, so that's better. And then all the values are higher than the deep seq R1 value. So that's the point of this release, right? Like stop using deep seq. We have a better model now. That's what OpenAI did here. Okay, so that's a recommendation for the paid users. Now, what about the free users? Because this is the most important one, I think, because it's going to be affecting the largest amount of people worldwide. What if you're not willing to spend any money on an AI assistant and you just want the best model out there? Well, I think, and according to these numbers, I would say that's also why OpenAI released it, that you should be using O3 mini in the free plan, as everybody has that in the free plan now. They released it to the public. They would have probably never done this if deepseek didn't happen last week. But all Free Mini is on the medium setting. And as you can see on the medium setting, we're comparing 80 to 80, right? On math, science, 77 to 72. 03 mini medium wins code. 96 to 96. It's a tie. And on software engineering bench, actually that's the one where Deepseek R1 performs better. So, you know, if you care about that one specific benchmark, then you should be using DeepSeek R1. But the time it took only 32 seconds for it to generate the result. Or on average, Deepseek took an average of 95 seconds, three times as much. And due to that, it's superior performance on most benchmarks and a way, way better time. O3 mini on the free ChatGPT plan is what you want to be using moving forward if you want the best and smartest AI. Plus, ChatGPT has obviously more tooling, custom instructions and the file attachments and stuff like that. So it's just a superior product in that sense too. And if you want the smartest model, O3 mini, on the medium setting or in the interface, this One right here, O3 mini, is the best model you could be using today. So basically ChatGPT took back the crown and that's what happened here. So I hope this helped. I have one final point, and the final point is many people keep asking Igor, like, okay, we have these new reasoning models. Like, they seem really potent. They are really potent. I'm using them. Amazing. What do I use them for though? Does this open up some new use cases? Because at the end of the day, like I try to make this focus a channel focused on all the AI tools for consumers and how you can make your life better with them. And there is specific things that these models do better than the previous generation of AI models. As many people say these days, you know, the GPT4OS and the sonnets 3.5s and then the llamas, etc. What they do really well and is they, they're good at planning, they're good at bigger picture thinking. We're not going to talk about that in this video, but I'm going to refer you to two resources that I already created, matter of fact. One is an interactive and open and fully free community challenge that we're running right now that I want to briefly show you. I think this is like one of the most amazing things we do here at the AI Advantage in the public area of our community. We basically have this space right here so you can freely access this if you just, you don't even need an account. You can just look at this and entire area. Look, I'll just close all of this and down here you have the public challenge. And the public challenge basically asks people, all of January viewers of this channel, like, what is your best use case for 01 or now 03 or Deepseek doesn't matter. And then multiple community members shared their best use case here. There's some really amazing ones. Look at that. Like splitting bills and having a financial advisor or turning order or Deep Seq into your personalized science research lab. We looked at this one in the last video. Really amazing. With downloadable files to run this all and a discussion underneath of people or using it or revolutionizing HR with these new models. All of these different approaches and more for you to view. And I'll be doing a public challenge on February 3rd, a public stream, I wanted to say. And on February 3rd we're going to be looking at all of the results here and we're going to be talking more in depth about the use cases. Plus there is one more video on the channel where I went a little more in depth on 01. I'll just pull this up right here. I believe it's this. It's this one. I'll link it in the description below. There's a video here that is all about. That is all about using. Oh, free. Okay, I'll just use it. I'll just link it below. So there you go. Use cases from the community. Separate video where I share all my thoughts. It's really good at translating rewriting and business activities that require more planning. Those are basically the three big use cases for me. If you're a paid user, use O3 Mini High. If you're a free user, use ChatGPT O3 Mini. And that's going to give you the best model in the world. All right. I hope this was helpful. Bit of a confusing release and I hope this brought you some clarity. Enjoy.