What is an AI Model? Definition, Labs and How to Pick One

Hey, my name is Mahan. I'm the CMO of Nairon. My name is Luka. I'm the CEO of Nairon. Luka, when Chad GBD first came out around 3 to four years ago, it was pretty much a category of one. Definitely. Now, fast forward 3 to four years, we have all these new models on the market.

You have Claude, you have Gemini, you have Chinese models like Deepseek. So in this video, break down for us exactly what an AI model is and what should people you know know in order to make the right decision on what to use, when to use it.

So essentially the model is the engine behind some of our favorite tools like Chach PT for example. I think it's important to distill this down that the model is the intelligence layer behind all of these tools, right? It's where a lot of the thinking happens. It's where the actual retrieving of information is happening, right? And our favorite tools like ChatGPT, like claude, like cursor and perplexity that we have here essentially are just built on top of the models themselves. So for example, ChatGPT is an app that's built on top of the GPT5 model for example and claude the app the co-working space that we that we like to use on our desktops that is built on top of the uh model itself, right? The cloud model, Anthropics model and then we have other rappers that are built by other companies. So not built by the actual labs themselves like Anthropic and OpenAI but rather like Perplexity Cursor. These guys are essentially rappers. They've built a product on top of the model themselves.

And rappers are things we'll get into in the next episode as well. For the time being, I think it's just to simplify it. It's a company that uses the model from OpenAI Anthropic in order to make another more nuance product.

Yeah, definitely. And so if we look at what all of the model providers are like what the landscape is, you have really five big players here. You have OpenAI. They are essentially the leader in the space. As you can see over here, we have three different models. We have the GPT5, we have the GPT5 mini and the O series. The GPT5 essentially is the smartest, largest model. The mini is a smaller model. And we'll get into why there are different models per company.

Um, but when looking at the O series, the O series is actually their open- source model, right? So essentially you can run the model on your own machine. And so looking at the anthropic models, you obviously know Opus. They also have the Sonnet and Haiku models. Those are smaller models. At Google, they have the Gemini Pro and the Gemini Flash, a very smart model, but they're also a very fast model. And then at Meta, they're more open- source. They try to do things that's available for any engineer to run and any company to run. And so they have the the O Llama and the Llama series.

And then you have Grock, right? It's a up and cominging AI lab. They're not used as much as the other four, but still an upandcomer in the space. And so why wouldn't just someone always use the most powerful model? Why do these tiers even exist?

So the tiers exist because the larger, more smarter models are slower and more expensive. And so if you have a very simple task like you want to clear up a transcript or you just want to identify a yes or no question from a simple block of text, you would most likely use the cheaper models for that type of stuff.

But if you wanted very complex thinking, a complex answer to a plethora of data sources coming in, you would use the biggest model. So here we're looking at anthropics three models, haiku being the smallest one and opus being the largest one. A lot of people actually use Sonnet as like the daily driver and also the biggest difference is the token usage drastically varies between all three of them.

And so what kind of mental framework should I use when picking out a model for a specific task? So the mental model that I like to have is this triangle where you either choose two out of the three um bases of the triangle. you either have the capability or how smart a model is with how expensive it is, but you lose out on speed or you look at a cheap and fast model, but it's not, you know, that that smart. And so you can really only have two out of the three available options.

And so when thinking about your task, we like to think about how much intelligence is needed, how quick does this operation need to get done, and how much budget can I set for the specific task. Obviously with how fast the space is advancing, you're seeing all these companies release new models, constantly compete with each other. So how can the average consumer know exactly which model is the best at the specific task that are looking to get done? So in order for you to really uh understand which model to use for which use case, you would essentially need to look at like a leaderboard. And so we love to use arena.ai. It's essentially a table similar to this one. This is a simplified version, but it essentially shows you in writing these models are out competing and they're done by peer-reviewed literature. So, a bunch of testers are voting on, you know, what has the best outcomes. In general, the models are pretty good and perhaps we don't need to really stress too much about this. This was the case more about 2 years ago where you really had to know this because there was such a big difference between the models. And so the larger labs are essentially building tools on top of the models to differentiate themselves rather than the intelligence of the models themselves like rappers of course.

And so guys, we'll cover rappers much more in depth in the next video. In the meantime, you can subscribe to our newsletter links down in the description below. You can also find us on LinkedIn where we're very very active. And uh we'll see you next time.