How GPT 4.1 Changes Everything for Developers

Let's say find the latest news and uh about productivity apps and respond with bullet points. And then from here like we had we could do 41, we could do 41 mini, 41 nano. It's all about pricing at that point and level of capability. So we can do the research and then the output with a different one. 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. Okay, so Chat GPT has been releasing models like crazy and one of them that just got released is a new model that can drastically change how you interact with the

API. So, we're going to dive into what exactly it is and how you can utilize it today. Let's get right into it. So, first and foremost, it is the chat GPT 4.1 model that is actually only available in the API. That's right, you heard that correctly. It's only available in the API. It is only, to my knowledge, the first time that this has been a thing, or at least it's been like a recent release where this has been a thing. As you can see, basically these are new models, 4.1, 4.1 mini, and 4.1 Nano. And these models outperform 40 and 40 mini across the board with major gains in things like coding and instructions. They also have larger context windows supporting up to 1 million tokens of context and are better to use with that context with improved long text comprehension. So they feature a

refreshed knowledge cutoff of June 2024, which is nice, too, cuz obviously there's like outdated data in the models themselves, which is kind of funny. As time goes on, they don't like get more info. They got to get somehow that stuff to pipe in automatically. Oh, wait. There's search features now with API and chat GBT. So, we're going to dive into all this fun stuff, but as you can see, some of the improvements are that the coding score went up a decent amount. It says it improved by 21.4%. And it's even better than 4.5 in some cases. Instruction following, which is interesting. 4.1 score is 38.4%, which is a 10.5% increase over 40. By the way, some of these improvements aren't going to be drastic. It's just a matter of the fact that they want to release it and release after release after release, these things

add up like iPhone improvements. Okay, long context for the long multimodal context understanding 4.1 sets a new state-of-the-art result, scoring 72% on the long no subtitles category, a 6.7 improvement over 40. So we can see right here this nano one is actually the most interesting to me and the fact that small model performance is getting better because from a pricing standpoint this makes everything more I guess I would say available from a pricing standpoint you'll see that coding has a significant increase here we see these benchmarks and versus 40 I mean this is just much better for verified accuracy there's a ridiculous amount of charts and showcases from the product here what's really cool to me is that it has improved front-end coding and is capable of creating web apps that are more functional and aesthetically pleasing. So, it showcases the difference between the two

that it made. And as you'll notice, this is the first time a small model it gets closer to these higher models for evaluation accuracy, multi-challenge accuracy, and other items where the smaller model usually is lacking much more than the 01 high or 03, 04, all of those different ones. 4.1 even nano is doing pretty well here and obviously the longer context window is something that we have to care about and I like the fact that from a multimodal standpoint it's improving too as these smaller models improve this is what's going to be crazy okay if it's better than 40 and it's a nano model you got to think about it like this take a look at the pricing right this pricing versus this pricing is pretty significant right so for basic tasks we have extremely costefficient models now. I've never seen a Nano before. This

is the first time they've released it, right? So, take a look at this. And the pricing of 40 was worse and now we have a model that's essentially similar for a cheaper price. And 41 Mini, once again, an improvement. Cheap price. It's crazy to me how cheap these are getting. I really wish we had these before, but hey, I can't complain. It just keeps getting better. And obviously, you'll you'll see the pricing difference at a significant level as these things scale and will continue to decrease. So we can dive into the showcase of this real quick. I'm just going to go into a scenario and make.com and I'm going to go to OpenAI and you'll see that there is an option to create a completion and in this list now you'll see there is you'll have to find it specifically. There's still the 40 option

in the API and you'll have 4.1 as an option. Now, something that is frustrating is in like a make.com situation, you don't actually have the ability to use like the 40 search, which essentially is able to utilize Google search, right? So, that's one of the critiques I have. Okay, if you look at 40, it's got some of these nice other options. I really want 4.1 to have the search capabilities. So if I go here to 4.1, I go here to assistant, I could add any text context, like write a social media post on the latest news about productivity apps. And what I can do is have the module before it be something that's pretty cool. Just have it literally be find news. Let's find news. And we'll just have this say write a social media post and latest news about productivity apps. And we can

say, yep, find news on latest productivity apps updates. Change this to the search capability in 40, right? And then we could literally have it say find the latest news and uh about productivity apps and respond with bullet points. And then from here, like we had, we could do 41, we could do 41 output with a different one. I really like the idea of nano. Reason being fact of the matter is most of these tools as they get better and you use the minimal pricing, it's going to be much better for you to scale whatever tasks you're doing and have more employees or people utilizing these things. So overall, it's a pretty solid breakdown, right? We have some of these news here and then it takes it and turns it into a social post. And at the end of the day, these are all going to

be pretty similar and just marginally improved. I don't really want to get into the nitty-gritty of like how the text looks at this point because it's just an improved 40 model at the end of the day for a solid price. I found the page, by the way, for detailed pricing. Here we go. Here we go. Here we go. So, we got 4.1, right? Okay. So, 4.1 is somehow cheaper, which is hilarious. Now, it doesn't have the search capabilities in that text, but it's fine. Let's take a look at 40 Nano. So, 41 mini is more expensive than 40 Mini. Okay. 41 Nano though is cheaper than 40 Mini. Interesting. Still the most expensive one for no reason is 4.5. And obviously, we have 01 Pro. That's really expensive. But for all intents and purposes, I do want to call one thing out. If we go

to the API here and we go to web search, you still can do web search. You can utilize a custom API call by building this out yourself using the what's called chat GPT create a completion here. What we can do is so we can do API call, right? And what this is going to be is we're just going to copy this header here and we're going to do post and what we're going to do is we're going to put into this body. We're going to take this v1 responses. We're just going to leave it like this. We're going to take this. We're going to use this as an example 4.1. And I just pressed enter and rightclicked. And you'll see if we just do a little search for the response, you'll see the model ran. And if I type result out and you'll see over

here, we can find text under the content here, which obviously you can put into an iterator or something like that. But text, let's change the prompt to be all-encompassing. So, okay, actually, what was the positive news from today? That's fine. Let's just have this be what showcase. I just want to show you how you can use 4.1 or whatever it is. And you can just change the name of the model that was here right in the API whether that be 4.1 nano like all you have to do is find the model names. So if you go to models you would find the model you want. So if you go to 4.1 you could search like 4.1 mini or nano right and then the name that you'd put in the API as you can see is this. So I could always change it to nano by

going here and just changing it to the name of the uh proper model. All right. And now if we press run. Okay. Actually it seems like 4.1 nano is not supported. So I apologize for that mistake. And it's good that I called this out. So 4.1 is actually available with web search but 4.1 nano not so much. Let's check about mini just to do a test. 4.1 mini seems to be available though. So 4.1 nano not being available with search is kind of frustrating but 4.1 mini that's fine. It's an improvement. was a positive news from today. Okay, so it says Scotland's reing efforts lead to significant increases in native woodlands, Denmark to ban phones to schools. Beavers legally released into the wild in England. Very cool. So overall, this is your 4.1. It's an improved, cheaper model that has much better handling of context

and coding that can be utilized only in the API. That's why we dealt with in the API, not in the front end. So if you want to try it out, make sure to go check out it at OpenAI. And if you like this episode, make sure to leave us a like, review the podcast, and we'll see you in the next one. Thanks for watching. Bye.