open AI just released their new 03 series for chat GPT we now have models available today that we can start using in testing so we're going to dive in take a look at some of the updates and we're going to give it a couple of prompts in order to test it out by the way if you love Ai and you want to keep this discussion going after the video about O3 miniseries things like deep seek and other tools then I highly recommend joining my free AI Community which I'll leave in the description and top pinned
comment below let's check out 03 so let's take a look at this open aai 03 Mini model series so right off the bat open AI starts this by saying the 03 mini is the newest most cost-efficient model in their reasoning series available in both chat GPT and the API today so it's stronger in things like science math and coding it costs less and takes less time than open ai1 mini that's very cool it says open ai3 is our first small reasoning model that supports highly requested developer features like function calling structured out puts and developer
messages making it production ready out of the gate for developers this is huge now here I think it's interesting it says developers can choose between three reasoning effort options low medium and high and this is going to allow you to optimize for your specific use case it allows O3 mini to think harder when tackling complex challenges or prioritize speed when the latency is a concern they did some evaluations by expert testers and it says the testers preferred 03 minis responses to 01 Minis 56% of the time and observed a 39% reduction in major errors on
difficult real world questions here's how it's performing in the competition math benchmark as you can see uh o 03 mini High actually scored 87.3 compared to even the 01 model which is their General broad reasoning model uh for accuracy and that got an 83.3 on the competition math so even the miniseries in 03 is outperforming the main series of 01 and PhD level science questions 03 mini high is outperforming 01 and 01 preview as well not by a lot but it still is outperforming those and this is just a mini model keep that in mind
they haven't released just the regular 03 yet and then in competition code pretty much every one of these 03 models are scoring above all the 01 series not even just compared to the mini series but 03 mini low is still behind the regular 01 model but both of the other medium and high models for 03 mini are scoring better in coding benchmarks than the regular 01 model and right here it says in AB testing 03 mini delivered responses 24% faster than 01 mini with an average response time of 7.7 seconds compared to 10.16 seconds so
as you can see the latency from 03 mini to 01 mini is actually 24% faster which is cool because it's also scoring higher in the benchmarks so not only is it faster but it's also smarter as well it says 03 mini is pushing the boundaries of cost cost effective intelligence by open Ai and they're definitely needing cost effective intelligence with deep seek on the horizon and uh how cost effective that is they uh definitely need this but it says this model continues to track our record of driving down the cost of intelligence reducing per token
pricing by 95% since launching GPT 4 while maintaining top tier reasoning capabilities so within our model selector we have new options we have 03 mini and then 03 mini high and it says this is fast at Advanced reason and O3 Min high is just great at coding and logic with both of these I want you to notice that you can search the web which is very cool because in 01 you can't even search the web and we saw that this model 03 mini high was actually scoring better on the benchmarks now you can attach files
yet but it is kind of cool that you can search the web and have live search with 03 mini high and just regular 03 mini so let me go to 03 mini and let me try scraping something maybe I'll just say scrape the N documentation and tell me everything I need to know about the HTTP request node now when you're doing something like building agents in an application like maker n8n or whatever you're doing it's really good to have documentation and an advanced reasoning model all at the same time like being able to search for
live what is is this application all about when you're building something so the fact that we have a high reasoning model within the chat GPT interface that has the ability to search usually if I'm building like an automation on n8n I would go to perplexity if I ever had a question because it has that live data and 01 did not so now we have 03 mini with search available so let me try scraping some documentation uh to see if I can get some information about a certain node I'm going to send that off and it
thought for a second I mean that was pretty quick and I believe we're going to be able to click in here it's still uh reasoning I guess but we can click in here and we can see what it's doing uh behind my head here and actually uh watch it live so now it's getting a comprehensive summary of everything I need to know about the HTTP request node in nadn as documented in the nadn docs so keep in mind this reasoning model is going to be able to help you with these problems a lot more and
it's giving me pretty much you know everything that the documentation says about it within NN so this is like a live website of an N documentation and I think it's really cool that we can now bring this in with an advanced reasoning model and we don't have to use something like GPT 40 right we can actually use one of these top tier models that outperform 01 in some cases because within your prompts like context is everything so being able to have a have a model that can reason at such a high level and be able
to pull in context from anywhere on the internet that is massive so next let's use the o03 mini High model which says it's great at coding in logic along with search activated and what I want to do is I want to create an Automation in Nan where it pulls information from my aura ring about my activity data so I have this ring that tracks my activity all day long and what I do know is this data when I do pull it into nadn is it's not structured in the way that I want it to be
so let's see if it can actually write us code in order to structure it into a Json like format the exact way that I need it and let's see how long it takes for it to solve this problem for me so as I was saying in this software nadn I'm building an automation that can track all of my activity and I want to store all of this activity within a Json file in my Google Drive that way I can chart it later down the line but as you can see this is how data is coming
in right now and I want to be able to extract all of this and then put it into a Json file but I want it structured in a certain way right now it's structured you know all over the place there's thousands of lines of Json here well not thousands but around 1500 so I want to see what can I do with this data for if I do have it in my Google Drive and a Json file and also can it create me code in order to accomplish this so so we are going to push this
model to its limits we're currently on the GPT 03 mini high and I've typed out a massive prompt and given it so many actions to complete let's see if it can help us with our actual real world automation coding problem because again I don't know how to code but I love using large language models along with search functionality and I'm very excited to see how this helps out because if we have something that can code for us and it can consistently quickly get things right for you then very powerful especially for low code no code
people but if you are a programmer then this is just going to be for Speed at this point right if you know how to ask the right questions give the right context then obviously it's going to be a lot more helpful but let's take a look at my prompt that we're going to test out I say I am creating an Automation in nadn that transforms my aura ring data remember my aura ring is that thing that tracks my activity in my sleep from activities into daily Json files in order to query with a rag system
system later down the road So eventually I'm going to upload all my activity data to a rag database I'm already in the process of doing that with my sleep and I want to be able to talk to a chatbot about you know what was my heart rate at this time of day or what was my heart rate at 234 last night you know I'm going to be able to do that with all of this so I say I want this data for each day I'm also creating Json files for my Google Drive in order to
chart and graph these stats as well so a lot of things already going on I say I want you to to do a couple of things and here's where I'm really going to test the reasoning of this model I say first review the aing documentation and this is the beautiful thing about having it be able to search the webs you can tell it to review certain documentation for you that it's going to need is context and then I gave the link to that ordering Cloud documentation and then I say in order to see how activity
data is structured and what each piece of data means that is coming in so I wanted to see within the documentation okay how is this data actually coming in is it useful does it matter and step number two I say review the n8n code node documentation and then I give a link to what that is that looks like this so within n8n this automation software you have different pieces you can put together and within their documentation they explain how those pieces link together so I wanted to review this documentation because not only is that going
to give it more context into my problem but it's also going to give me up to-date it's also going to give it upto-date information on how to help me solve my problem right so I said review that in order to understand how to use this node to transform my current data and structure Json data and then I say step three review the current structure of data being pulled in with my workflow that I just uploaded step four so I'm giving a ton of different steps give me suggestions for flat Json files that I can create
from activity that would allow me uh that would Aller cool graphs with met minutes so met minutes are just like activity minutes every 5 minutes it tracks what your activity levels are with the ordering so I gave that as an example and any other things that I could create in sightful graphs from with Json files later down the line then I give one last step I say give me the code I need in order to capture full raw Json data organized nicely that goes from my set node so I'm giving it context into the labels
of my nodes in n8n this is called a set node you basically can pull in data and then give it a name here so I'm basically pulling in all of my aing activity and then just labeling it Aura data so it's in a nice uh array in order for the code node to better use it so I'm telling it to you know I'm giving it context into what type of node I am using each one of these boxes are called a node and then I say here's how the data is coming into the code node
so with this this outputs certain data as you can see over here on the right it outputs data that looks like this so I mean and it's uh like we said 1,500 lines long or more so what we want to be able to do is pull in that data to um that code node and then have it get in this more structured Json uh data that we can then download and convert into uh binary so I give it that entire structure okay and it's super long I mean you can see my scroll bar how long
this prompt is and how much context and things I've given it over here on the side but then at the very end here I just say complete all of my steps in order to help me out so with all that being said I think I can send this off and remember the Json that I uploaded although it just has a lot of these little numbers it's still 1500 lines of uh Json data that's raw so let's see how it does with this I'm going to send this off I'm going to go down let's see if
it's reasoning through this this it looks like it is if we click over here we can see all of its thought process that's actually going into it it's enhancing the Json structure crafting the code already so it's very cool what this thing can do hopefully it works I'm interested to see does it actually work and even more interesting than the code I'm interested to see uh what type of ideas is this thing going to give me because remember up here if I can get to the top of my prompt I asked to give me suggestions
for flatten Json files I can create from activity so like as you can see I mean it's going to work in here it's like it's really thinking through before it gives me an answer look at all these things it's thinking about over here on the right hand side of my screen so I'm interested in suggestions right what is it going to interpret for my data okay so it gave me an answer it only had to think for 13 seconds but it thought about quite a bit of different things over here I can close that it
says overview of the aing activity data and then it says according to the aura API documentation it kind of goes into each one of the fields that I could pull in so this is good because if it's typing this out and telling me this it understands it itself right it's using this as context now so it knows all of the little uh parameters that go into making up activity data it covers the nadn code node capabilities it knows which node I'm trying to use and it's referencing the documentation right here and then it does an
analysis of my current data structure it says your activity raw node outputs data that looks like this and then this is what I was most exciting about it gives me suggestions for flat and Json files it says for insightful visualizations you might want to create a daily summary Json that flattens the data for example a flattened Json could include uh metrics such as score steps activity calories total calories Target calories Target meters date and time stamp activity breakdown and then contributors these are the things that I'm really interested in and then it gives me different
kind of graphs I could create with my data you know I gave it the context of my data and now it's saying like it's giving me all these suggestions of different graphs I can create and that understands which type of metrics I can pull in with the documentation that I told it to search and then it gives me the code right here that I can use in order to uh get all of these files that I need so I'm just going to try this out this it's going to give me the full raw all activity
data so I'm going to copy this I'm going to head back to nadn then I'm going to upload a code node so I'm going to type in code and now I'm just going to connect it well it's already connected which is good and now within this code right here I can just paste this in let's see if this works now it's not much but like instead of trying to figure all this out myself I just upload what my data looks like tell it what I want it to do it gives me suggestions for what I
could do and it would be cool if this worked and output uh more structured Json so let's see if it does I'm going to hit test step and let's see if it works beautiful it looks like it worked first try and usually it's not like that you know usually I've got to iterate I've got to go through multiple different errors and it doesn't actually output my data how I want it to but now it's giving me the full raw data organized by day so it's really organizing and structuring my data the exact way that I
want this is beautiful so now I could even come in here I could you know convert this file to an actual Json and that's easy because now that I have the data all mapped in here that's all I've got to do is just map it to the code node and so now if I test this step it's going to convert that into a Json file which I could then upload very easily to my Google Drive so this has been an example of using the 03 Mini model in action for a problem that I was just
about to go through and actually use 01 for I actually had the 01 model selected and then I got a notification that said 03 mini was available so I hope that you enjoyed this video kind of covering the update if you want to stay connected and you want to ask me questions about 03 you want to chat about 03 together then I recommend joining our free AI Community AI Pioneers now ai Pioneers is a place where you can just come chat about AI with other people we have 36 people online right now over 3,400 members
in the group and it's kind of just a fun spot to be about AI you know we have people hiring in here we have people just talking uploading valuable content so great opportunity to network with people give your own advice uh ask questions you can go to the classroom we have all of our YouTube videos in here with a 10 module AI course as well so if you love Ai and you want to be with a group of people who also love it AI Pioneers is a spot to be but with that being said I
hope you enjoyed this video with 03 if you did please like And subscribe and I'll see you in the next one
