NEW GPT 5.5 LEAKS are INSANE! 🤯

So there's been some brand new leaks on GPT 5.5 now people are saying this is coming anytime soon a lot of people saying they expect GPT 5.5 itself to be released today or this week minimum now this has not been confirmed by chatGPT but these are new leaks new situations i'm going to walk you through some examples so there's a bunch of new models that have been leaked including Heisenberg in Glacier Alpha GPT 5.5 and a new one called OAI 2.1 now if you're not familiar with any of those that's because you know most of the world is not familiar with each of these right but basically as you can see in the screenshot by testing catalog who kind of like focus on finding new AI models and that sort of thing they were saying that basically GPT 5.5 and a bunch of internal model names have been spotted on codecs very recently and this normally signals that OpenAI is preparing a new model okay so we've got for example OAI 2.1 we got Arcanine we got Glacier Alpha we got Glacier Alpha Block CY4 and Glacier Alpha Block CY3 now this was leaked as you can see right here from OpenAI codecs this is a screenshot from codecs and these are the new models that you can select right now this was actually pulled within a few minutes but basically you can see another example here and this this was accidentally leaked and these are the internal models that were accidentally put out there as you can see in this tweet by Kan and this was only for pro accounts right so you can see this drop down here as well so it was spotted inside codecs and it was spotted inside chat GPT directly and so pretty crazy stuff so let's come on to like what this means what's happened etc what the differences are you know how this has happened all right so normally when testing catalog announces stuff like it comes out pretty quickly so you know pretty good stuff right there let's talk about what it means so on April 22nd some codecs users on pro accounts opened the model selection drop down and saw a list of internal model names that were never meant to be public right you can see the examples right here it was a UI level exposure an internal testing environment that accidentally got pushed to production so OpenAI pulled access pretty quickly but not before developers screenshotted everything and some even managed to run prompts on it now from what people were saying they were actually saying that this is like much better it fixed a bunch of issues now there's basically a bunch of new models and i'm going to run you through exactly what they mean which one each one is etc so there's GPT 5.5 right this is supposed to be their latest frontier agentic coding model this would be one right it's code named spud and it's designed for pre-training it's already done on pre-training right pre-training completed March 24th Greg Brockman from the OpenAI team actually confirmed it and said there's two years of research inside it right Sam Altman called it a very strong model indeed that could totally change the economy and some developers actually got prompts in on it before it was pulled yesterday and once it actually fixed the problem in three minutes it's been stuck for three hours for four hours now listen the these things get over exaggerated so you know with me particularly with OpenAI releases i don't take them like that seriously i love testing this stuff i love testing for example the new agent workspace stuff that came out today i loved testing out image 2.0 yesterday amazing but at the same time i take it all with a pinch of salt when it comes to some of the marketing hype that comes out there you know what i mean so obviously it said it felt much faster way more token efficient and a clear setup for front-end development so that is the GPT 5 model that got released and then there was some other stuff as well like you can see right here so there's OA 2.1 right and this is the described so these are like the descriptions of the models right the internal descriptions of the models now the name is an internal name so OA 2.1 basically this has the same description as GPT 5.5 OA 2.1 so same description but a completely different naming convention which is interesting because it doesn't follow the GPT naming pattern at all so this one you know OA 2.1 it could be for example OpenAI testing a new naming system for their models it could just be a lightweight or specialized variant right nobody knows i'll give you an example so what this could be is it could be like an advanced model it could be a different model that's split testing or what it could also be is a model that is fine-tuned for something specific that we don't understand yet right so that's another option as well and then we have GPT 5.4 obviously that's pretty standard nothing to worry GPT 5.2 codecs we're not really too worried about that we have Arcanine right this is super interesting so this is a frontier model these are the descriptions by the way by OpenAI a frontier model with legendary appetite for starches right whatever that means you can see another example here of the leak so someone found the model and then posted it on reddit essentially and these are the new models so Arcanine is a frontier model with legendary appetite right it's obviously a playful sort of internal description i think it's named after the pokemon some people saying and some people think it could be targeted at specific logic tests or benchmarks these are just like people's interpretations what it could be people's best guesses right others think it might be a specialized model for a particular domain right again like a fine-tuned model the fun description suggests it might be an internal experiment or research model right bear in mind like for example they called agent workspaces that just dropped if you haven't if you're not familiar with agent workspaces this is agent workspaces right so you go into chat GPT make sure you're on a business account then you're gonna scroll down you see these agents over here right and these are free to use until May the 6th and so Arcanine could be could just be a frontier model you know for example Hermes this was called Hermes agent originally that's what agents was co-named so they always have like different names uh depending on what the situation is and depending on you know what they call it internally then we have Glacier Alpha right intelligence that moves continents this one has history so Glacier Alpha was previously known as a codename for GPT 5.5 itself which is pretty interesting but in this leak it shows up as a separate model with its own entry so either it's a different checkpoint of the same base model or it's evolved into something distinct from what shifts as GPT 5.5 so there's a lot of speculation that leans toward it being related to long context or um stateful AI systems one sec um we also have Glacier Alpha block CY3 ice cold intelligence right so this is another um example right here right so ice cold intelligence a variant of Glacier Alpha the block CY3 suffix suggests this is a specific checkpoint or training run so you know like you have different training runs you have different checkpoints for different AI models if that makes sense right so this these could all be like versions of chat GPT all different stages of the same model being tested against each other right which is pretty normal in like AI labs when you save different versions during training and compare them and then you've got Glacier Alpha block CY4 ice cold intelligence as well so this has the same description as CY3 again this could be another checkpoint variant of GPT 5.5 but the fact that there are multiple Glacier Alpha variants suggests OpenAI is actively iterating on whatever the model is so they probably have like more advanced options you know it's not just about like one single option if that makes sense then you got Heisenberg um GPT Rosalind these were already officially announced recently right this was recent stuff but basically what does this tell us well it tells you a few things right number one OpenAI's internal model development is way more extensive than what they actually show publicly so eight or more unreleased models suggest they're running parallel research tracks you know general coding science long context lightweight variants all at once the shift to agentic is real here too so almost every model description includes the word agentic right these aren't for example just back and forth inside chat GPT they're models built to plan tasks implement multi-step workflows operate autonomously bear in mind like their entire research direction has moved towards AI agents they released this new option right here and also they recently partnered with OpenCLR I think they actually bought OpenCLR didn't they yeah so in you know in February 2026 OpenAI actually acquired OpenCLR right and then you got domain specific models as well right so for example Heisenberg and that sort of thing Rosalind for science research a lot of cool different models being released here so what do we know here well GPT specifically 5.5 we know its code name is SPUD we know pre-training is already completed we knew it was in safety evaluation and now it's in final testing um Greg Brockman has said two years of research inside the model Sam Altman said very strong model that could accelerate the economy um and also he actually said on the 22nd of um this is something interesting I don't know if you saw this right so basically he told people to come to the light side on x and I'll tell you what that means in a second so someone was complaining that uh that Anthropic right had leaked too much um and basically someone Sam Altman said come to the light side when people are complaining about Anthropic removing clawed code from the pro plan right so that's what happened there and then this was just yesterday and then someone said if you release GPT 5.5 or 6 on Thursday I'm on my way right this is what Dennis said and then you know we got the emoji the salute emoji that basically suggests GPT 5.5 could be coming so who knows but that was interesting in itself early test of feedback is that it's much faster more token efficient major step up in front end development solve problems in minutes it took hours on GPT 5.4 again that could be exaggerations and don't quote me on that don't go crazy on that like just see what you think um and the official release name might not even be GPT 5.5 right like open air hasn't committed so if the performance leap is big enough they could skip straight to GPT 6 although I have a feeling that it will be GPT 5.5 like I just I can't see how they're going to pull out a super powerful model from nowhere right um so let's talk about the context as well so anthropic had some situations with leaks and claude mythos this week I've done another video about that but basically this is pretty exciting stuff I would expect the GPT 5.5 is coming out anytime soon and who knows when that's going to happen but I would expect that it's coming out by um yeah within I would I would expect genuinely that it's coming out by Thursday that's what I expected so thanks so much for watching if you want to get more training and stuff like this feel free to get it inside the AI profit boardroom link in the comments description or just go to the aiprofitboard.com inside there we have loads of people using chatgpt using claude using AI agents if you like this sort of stuff you can get really practical tutorials on exactly how to implement it inside your business inside the community you can ask 2,900 members for help for support these are all people who are fully focused and really dedicated to learning AI and automating their business we do a weekly update where we break down all of the updates and we show you okay here's what to focus on here's what you can ignore so that you can stay laser focused and never feel overwhelmed again inside here as well we have a daily accountability group so that you never feel overwhelmed and you can just focus on your goals inside the calendar you can jump on weekly coaching calls get help and support in real time you can share your screen ask questions about your setup meet other people using AI agents and find out what other people are doing inside the map you can connect with people in your local city who are using AI agents like open claw like Hermes etc and then inside the classroom too you can get new advanced AI tutorials so for example we just did a full training on chatgpt workspace agents which was amazing and then also we have a full six-hour course on how to use open claw and a full two-hour course on how to use Hermes right so everything that you want and need to learn and win with AI agents is all inside the aiprofitboarding feel free to get it link in the comments description or just go to the aiprofitboard.com thanks for watching i'll see you on the next one cheers