GPT WTF?! OpenAI Model Madness Explained

The OpenAI Model Naming Confusion

Here's where it got super confusing. I mean, the task model was already confusing, but they added O3 Mini and the funny thing is they had to skip O2. So O3 is actually O2, but they had to skip O2 because there's actually an O2 Arena in London. They didn't want to confuse people with O2 Arena and probably couldn't do it because of copyright issues, so they had to call the next reasoning model O3.

You'll notice that I had to add numbers to my diagram because I was getting lost just trying to explain this, and I have a freaking diagram! So if you're confused, this is not something anyone—not even AI—can balance in their head.

When I Thought I Was Losing My Mind

When OpenAI released GPT 4.1 after 4.5, I genuinely thought I was losing my mind. And then they released O3 and O4 Mini and O4 Mini High. If you feel like I'm speaking a different language, that's because I am—I'm speaking OpenAI model naming speak, and I was dumbfounded.

So I thought it was time to revisit all of the OpenAI models and demystify this for myself and for everyone else, because it's kind of getting out of hand here.

A Trip Down Memory Lane

Let's take a quick trip down memory lane and look at the literal release sequence of OpenAI models. We started at 3.5 and then went to 4, and I was like, "Okay, that makes total sense. I'm on board." Then they went to 4O and I was like, "Okay, so O is Omni, alright, I guess that makes sense."

Then they reset and went to O1, and I was like, "Okay, so O1 is starting over, alright, let's progress." Then they went to O3, and the reason is because there's the O2 Arena, so they couldn't do O2 because that would cause too much confusion.

So this is why I don't think we're getting AGI, guys. I'm just kidding! Then O3 went to 4.5, so there are different flavors of ChatGPT. 4.5 is like a continuation of the GPT series, so 4 and then we go to 4.5. You can see I was already super confused.

Testing O3 for Research

You may have seen on Twitter that people are claiming O3 is AGI, and I really wanted to believe them. So I tried to use O3 for all the research tasks on this video. Unfortunately, I found myself going back to Perplexity and Claude for doing the research tasks. I found it was just way easier and more accurate.

How's that for irony? ChatGPT couldn't be an expert on itself! But I will say that O3 did some really cool things in the research tasks that I didn't see possible before.

Cool Things O3 Generated

When I wanted O3 to generate a timeline of all the different GPT models, it did a bunch of programming to visualize this and then exported an image. It wrote its own code and exported a timeline. When all the labels were squished together, I asked it to fix that overlap issue and it was able to generate it.

Something to note: when you generate these things, make sure to download them immediately. If you don't download immediately, it will fail—you'll see "Code Interpreter session expired," and you won't be able to get the full resolution version.

Where I Lost Trust with O3

I want to share a quick example of where I started to lose trust with O3. It was with a simple research task where I asked it to create a chronological table ascending by release date of the OpenAI models released starting from 3.5, including all variants.

I gave that to ChatGPT and to Perplexity with Claude reasoning. You would think that ChatGPT would know its own models, right? ChatGPT showed only 16 results, while Perplexity Claude had 21. ChatGPT was missing GPT 4.1 variants like mini and nano that were clearly present in the Perplexity results.

I was like, "Man, I can't even research its own models." So after that, I really started to lose faith in the AGI narrative and switched more of my research tasks back to Perplexity and Claude's reasoning.

Understanding the Model Lines

After further research, this is my best understanding of the models. I divided the AI models into three lines: Classic, Omni, and Reasoning. The dashed lines mean discontinued or going to be discontinued, and the solid lines are currently available.

Classic Models: These are text-only models. We went from 3.5 to 4.0, which made a lot of sense. From GPT4, we went to GPT 4O with the whole Scarlett Johansson controversy and advanced voice mode, which was amazing until it got taken away from us.

Omni Models: GPT4O was supposed to be omnimodal, so they released advanced voice mode and eventually camera integration. From GPT4O, they released O1 with chain of thought built in for reasoning.

Recent Confusion: After O1, we got GPT4O with tasks—the first model that can do scheduled tasks. Then O3 Mini, then GPT 4.5 (all about vibes and high emotional intelligence), then GPT 4O with image generation (the Ghibli craze), then GPT 4.1 with its massive 1 million token context window.

The Latest Models

Finally, we have the O3 full model and O4 mini model. These are the next level of reasoning plus tools. The big innovation is that not only are they smarter models, but they can actually use tools within the reasoning process. O1 could use tools, but it was after generating. The innovation here is O3 and O4 mini use tools within the reasoning—zooming into images, cropping, OCR text recognition, and programming all in the reasoning steps.

Making Sense of the ChatGPT Menu

Here's how to understand all the different models in your ChatGPT menu: GPT 4O is for Ghibli tasks and image generation, 4O with scheduled tasks is for automation, GPT 4.5 is all about vibes and writing, O3 is currently the smartest reasoning model by OpenAI, O4 Mini is the fastest reasoning model currently available, and O4 Mini High is the longest thinking fast reasoning model.

I hope this gives you a better understanding of the different OpenAI models. My freebie for you is this explanation as a PDF reference guide, available in the description below.