# ChatGPT o3 Mini is here - Best Model I've Ever Tested

## Метаданные

- **Канал:** Skill Leap AI
- **YouTube:** https://www.youtube.com/watch?v=0K1mljDik4A
- **Дата:** 31.01.2025
- **Длительность:** 9:47
- **Просмотры:** 60,810

## Описание

OpenAI o3-mini is here! A faster, more efficient AI model optimized for STEM reasoning, coding, and logical problem-solving. It offers adjustable reasoning levels, supports key developer features, and is now available to free users in ChatGPT. Plus and Team users get higher rate limits, and Pro users have unlimited access.

▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬
MORE FROM SKILL LEAP:
💡 Join the fastest-growing AI education platform & Instantly access 20+ top courses in AI:
👉 Start with a free trial: https://bit.ly/skill-leap
▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬

## Содержание

### [0:00](https://www.youtube.com/watch?v=0K1mljDik4A) Segment 1 (00:00 - 05:00)

Chad GPT just rolled out 03 mini their first reasoning model that is available to all users including free users I have it available here on my account they actually come in different versions of 03 mini that I'll show you here and kind of how they compare to the old 01 reasoning model but 03 mini right now is rolling out to every single version of chat GPT okay first let me show you what 03 mini is here from their blog post and then we'll test out some prompts here inside of my account with O3 mini and a couple of the different versions they have available right now and they pushed the release forward because of the pressures they got from Deep seeks R1 release that's a free reasoning model from a Chinese startup that got released about 10 12 days ago and it went completely viral so they decided to push this forward and it got just released right now today available to all users of chat GPT now right here the one thing they pointed out is this is exceptional at stem capabilities so that's math science coding things like that now it's also low cost for developers that are going to build applications on top of it the latency is lower and this is going to replace the 01 Mini model from the model's drop down so 03 mini is going to take the place of that now right here chat GPT plus team and pro users have access to 03 mini starting today and then down here it says starting today free plan users can also try 03 mini by selecting reason in the message composer this marks the first time a reason in model has been made available to free users of chat GPT which I think is a really big deal pretty much everyone that uses chat GPT for free has never even seen the capabilities of the 01 reasoning model or the 01 mini reasoning model and now they get 03 mini reasoning now this is interesting while open AI 01 remains our broader general knowledge reasoning model o03 mini provides a specialized alternative for technical domain requiring prec precision and speed so 01 is probably going to get replaced with 03 and that's going to be the general model and the mini is going to be more specialized for Speed and domain specific knowledge and this is also important to note all paid users will also have the option for selecting 03 mini high in the model picker for higher intelligence version that takes longer to generate responses Pro users will also have unlimited access to both 03 mini and 03 mini High so this is really interesting they have these models here based on the type of effort they put in so you could see low reasoning effort medium reasoning effort and high reasoning effort really strange how they name these models so you got 01 mini then low 01 mini medium 01 mini high in my account here I have 01 mini and high this is the pro accounts I still don't have it in my teams or the free account but these two right now are going to give me unlimited access as long as I have the pro account which is $200 a month and if you look at some of this Benchmark right here you could see 01 is at 83 01 mini high this is the best version of 01 mini it's beating the full-blown 01 version right here now when it comes to PhD science questions it's almost exactly tied with 01 here but it's a little bit shy of that here but probably 03 the full-blown version of it is just going to beat these two by quite a bit now I think this is where it really shines it says in this competition code look at the numbers here pretty much every version of it beats 01 preview by a quite a bit and these two versions also beat the full-blown 01 and a huge leap from 01 mini over here and here's a software engineering Benchmark over here again 03 mini high beats the fullblown 01 model as well now they have other benchmarks on this page too but they don't have anything where they compare it to other available models they're only comparing it to their own models there is nothing against deep seek R1 or gemini or things like that but I'll go ahead and make videos comparing it to those very soon too okay let's take this for a quick test drive here I'm going to start here with chat gpt1 mini over here and I actually just posted a video comparing 01 versus deep seek R1 the reasoning models and 01 failed two questions so I'm going to test out those questions first the first was which one is bigger 9. 11 or 9. 9 this is something that 01 failed when I tested it just yesterday so let's see how it Compares here that took four seconds let me click on the reasoning okay let's see comparing very little reasoning required and therefore 9. 91 is bigger than 9. 11 and again 01 got this wrong just yesterday when I tested it so this is a pass for this one and I'll start a new one here this one it also failed kind of

### [5:00](https://www.youtube.com/watch?v=0K1mljDik4A&t=300s) Segment 2 (05:00 - 09:00)

halfway it gave me half the question but this had a little bit of twist to it because I was looking for two viable answers a horse cost 50 chicken cost 20 gold cost 40 you bought four animals for a total of 140 which combination did you buy okay let's see what it does here let me click on this reasoning I want to kind of see it okay it has a little bit of thinking okay this is actually has more steps than 01 dead and it is showing its work over here too so that's nice let's see both combinations give you four animals for a total of 140 you could have bought animals in one of two ways zero horses one chicken three goats two horses two chickens zero goats okay perfect this is again another question in my test just yesterday deep seek R1 got it right 01 got it wrong and in the very first one this has got it exactly right and both sides of it the other or one model only gave me half the answer but this was more of a trick question to see if it gives me every possible I correct solution let me try 03 mini high and I'm going to actually give it a coding question because on this drop down it says great at coding and logic and this is something I never got any model to get right I'm going to ask it to create a game of chess for me I could run locally on my computer okay so here's the game he gave me the pieces are not there because that would require me to download them from the web so it just made these but I could kind of guess where things are this is a pawn so this moves this way they move this way this should be my queen that is the right way it should move just move this guy down okay this is my Bishop here I'll move it here let's say they move here and then this is my queen you can move this way yep all these are right and I should take this piece okay that should be Checkmate let's see yep it's Checkmate I didn't get a message but it's not letting me move any other pieces so knows the game is over wow that worked really well I'm going to try one other thing because one of the best updates in this is this right here which I was trying to actually use right now but let me try again I could now turn on search and let it combine this reasoning model with search so I'm going to ask it to find me the pieces that I need to make this look like an actual chess game instead of those shapes wow this is actually incredible right here CU we never had the option of combining search with a reasoning model inside of chat GPT it found a link for me I click the link it found these pieces for me and then it's saying save these to a folder called assets on the same directory okay I'm going to go ahead and do that and then run the game again oh wow we got ourselves a full-blown chess game let me try that again okay move this over here move this down here and oh wow that is just wild that worked perfectly I have never in two years of testing AI tools got usually not even a checker game to properly work all the way and this literally told me where to find all these pieces on a website copy and paste them make a folder that's incredible now here's another stem based question and I have no idea how to solve this but I do have the answer key and if you have harder questions I'm doing a deep dive video I'm going to compare this against deeps R1 so please let me know in the comment section what those are so I could test that out in the very next video okay I do have the answer key the answer is 12 so I got that one right too and I did not even have to use 03 mini High I just used the regular one that everybody has access to now okay here's another one again I have the answer if you want to pause it and try to solve it I again have no idea how to solve this okay and this also matches my answer key if you could double check that for me if you understand this kind of a problem let me know also if it's correct that's what my answer key had as well so far I can't even come up with questions to throw 03 mini off so I'm going to do some more digging to try to find harder questions I'll look at all your comments in the description and then I'll test that out the 03 mini versus 01 and deep seek R1 to see what we come up with but from my early testing it's extremely impressive extremely fast you got web search on top of reasoning it's available to free users inside of chat GPT so I'll do some more deep Dives I'll check out your comments for more suggestions and I'll catch you on the next video

---
*Источник: https://ekstraktznaniy.ru/video/12589*