Although the entire AI gravy was triggered by just oneChatGPT manikin , a band has changed since 2022 . novel model have been resign , old modelling have been replaced , update roll out and roll back again when they go incorrectly — the world of LLMs is pretty busy . At the second , we have six OpenAI LLMs to select from and , as both user and Sam Altman are aware , their name are completely useless .
Most people have probably just been using the newest model they can get their workforce on , but it turns out that each of the six current example is effective at different things — and OpenAI has finally decided totell uswhich model to employ for which tasks .
Why are there six models in the first place?
LLMs are irregular — users never have intercourse what form of responses they will get , and the developer do n’t really know either . certainly , it might be more convenient if we had all of the capacity uncommitted rolled up into one model , but that is n’t as easy as it sound .
As OpenAI tweak its models , some things get better andother thing get worse — and sometimes unexpected side effects happen . There ’s no relation how long it would take to equilibrate things out perfectly , so it makes more sensation to just issue new interlingual rendition even when improvement are only focalise on a few expanse .
The results of this approach are the six main models we have right now : GPT-4o , GPT-4.5 , OpenAI o4 - mini , OpenAI o4 - mini - in high spirits , OpenAI o3 , and OpenAI o1 pro mode . And I ’m just going to say it again — these name really are useless . OpenAI may have give us a written document explain what each one does now , but that does n’t think you ’ll be able to think which name matches which capabilities — so consider write this little bearded darnel rag from the document if you take to think .
OpenAI
GPT-4o
Part of the latest 4o family of models , GPT-4o “ excels at unremarkable tasks . ” This includes :
you may search the connection with it , give images , use modern voice features , study data , and create custom GPTs . you may also upload various file case to help your prompt .
consort to OpenAI ’s own research , however , 4o does have a moment of a delusion job . It ’s not the spoilt of the bunch , but it did hallucinate around doubly as much as o1 during testing .
This can be problematical if you ’re using it to search the web or get word new thing — the trickiest aspect of hallucinations is that they often fathom alone plausible , making it harder to just “ check when something fathom off . ” Instead , the only room to be sure is to checker just about everything that you do n’t already know to be on-key .
GPT-4.5
agree to OpenAI , GPT-4.5 ’s solid suit is emotional intelligence . This stand for it should be practiced at help you put across with other citizenry , with prescribed recommendation including :
With other strengths such as clear communicating and creativity , GPT-4.5 is better equip to help you find the perfect tone or phrasing for specific situations — and ensure everything still sounds human .
OpenAI o4-mini
One of the more terribly named models , o4 - mini drops the “ GPT ” element of the naming scheme and awkwardly switch the 4o around to o4 . It ’s a smaller fashion model , which means it ’s not stuffed to the lip with as much random internet information as a full - sized role model .
The upside of this is that it ’s warm and less expensive to run , and the downside is that the example has less “ world noesis ” and is prostrate to hallucinating to make up for that .
Instead of asking it questions about the world , OpenAI advocate using o4 - mini for fast technical labor . instance let in :
OpenAI o4-mini-high
Here ’s another fearful name when viewed in isolation , but middling easy to understand if you already know what OpenAI o4 - mini is . It ’s still a small model , but it ’s a step up from the normal o4 - mini because it “ thinks longer for eminent accuracy . ”
This makes it upright at more detailed coding tasks , math , and scientific explanations . Here are OpenAI ’s case :
OpenAI o3
This is technically an older model ( because it does n’t have a “ 4 ” ) , but because the o4/4o fellowship did n’t make improvements in every area , it ’s still very relevant . o3 is peculiarly good at complex , multi - step tasks — the kind of projection that need to be done in multiple point with multiple prompts .
This includes strategic provision , elaborated analyses , across-the-board code , ripe maths , science , and ocular reasoning . If you want to start a task that you know will take a multiple - straightaway session to finish , using o3 will serve minimize the chances of the modelling losing track of the setting or hallucinating halfway through .
OpenAI suggests use cases like :
OpenAI o1 pro mode
OpenAI o1 is now considered a “ bequest model , ” though it is n’t evena twelvemonth oldyet . The “ pro way ” version is tuned for complex reasoning — which means it have more prison term to consider , but in return give better thought - out responses .
o1 also gets the just scores on OpenAI ’s PersonQA evaluation , which measures the rate of delusion . During testing , o1 hallucinate around half as much as o3 and three times less than smaller model like 04 - miniskirt . If you ’re a big ChatGPT user and your sessions incline to move long , then minimizing the pace of hallucinations could save you a decorous chunk of prison term in the long run .
Here are OpenAI ’s examples :
How to use different ChatGPT models
Unfortunately , you may only access GPT-4o and GPT-4o miniskirt on OpenAI ’s gratuitous level . If you ’re a Plus , Pro , Team , or Enterprise user , you’re able to use the model selector tochoose which modelyou need to utilize .
ChatGPT is also integrate into various other third - political party products , both free and give , so it ’s deserving checking which models different product habituate . For example , my pay search locomotive , Kagi , gives me access to multiple OpenAI models . There are also muckle of other AI aggregate service out there that give you access to multiple models from OpenAI and other fellowship for a more affordable damage than subscribing to each company on an individual basis .
While this data about the different models is utile to have , it does n’t impact everyone . If you mostly use ChatGPT to generate image , search the World Wide Web , and send general interrogation , then the nonpayment GPT-4o is wholly hunky-dory . It ’s only if you ’re into computer programming , maths , science , or particularly large projects that you might want to cogitate about which model is best for the caper .
