GPT 5o — The New Agents Era is Here! Features EXPLAINED

Goodbye Google Chrome, Safari, and every other browser crowd in my doc. Ever since I moved my entire workflow into ChatGBT, I hardly ever open a search bar. And now with the new Asian feature, old school browsers are now basically history. It can now browse websites, write code, fill out forms, even whip up slide decks while you sit back. This isn't the all chatonly GBT. It thinks and acts on its own computer to get stuff done for you. When GBT lands this summer, we'll completely rethink how humans navigate the internet. Web search reimagined. I'm already eyeing my task list wondering, can I upload half of this to Chad GBT? Alman teased that GBT 5 will bring

significant improvements over GBT4. While he didn't spill all the details, OpenAI's team has hinted at some game-changing upgrades. One big change is how new models are named and updated. Alman noted that with GPT4, they quietly shipped many enhancements over time, some internally called GPT 4. 1 or GPT40, leading to confusion about versions. This time they're aiming for clarity. Ideally jumping straight to GBT 5 and later GBT6 as distinct major models rather than a bunch of halfstep names. — I hope we don't have to do that. I am excited to just get to GBT 5 and then GPT6. — In practice, this means GBT5 should absorb all the upgrades that might have been called GBT4. 5, etc., and launch as one big leap. It's not just a faster model. We're already seeing signs of how GBT5 integrates with the new agent system, letting it not only reason better, but also acts smarter. Instead of giving you a clever answer, it just gets the job done.

Perhaps the most exciting confirmed news, GBT will unify OpenAI's various model lines into one system. Right now, if you use Chat GBT, you have to choose between models like GBT4 versus code interpreter versus O series models. OpenAI says the confusion is ending. We are not only pushing a new boundary with GBT5, we are also uniting our two series, says Roma Huitted, head of developer experience at OpenAI. He explained that GBT5 combines the breakthroughs of the reasoning focused Oer 023 models with the multimodal prowess of the GBT series. Instead of having separate specialized AIs, GPT5 is one model to rule them all. a single AI that can handle text, images, audio, and more in one go. Jerry Torrek, an OpenAI VP, confirmed this unified vision, saying GBT 5 is our next foundational model meant to just make everything our models can currently do better and with less model switching. And now with the new chat GBT agent feature, that vision is already arriving ahead of GBT5. The agent can browse the web, run code, fill out forms on its own. GBT5 will likely make this even smoother, fusing deep reasoning with realworld action. Imagine asking GBT5 a tough multi-step question and it silently delegates subtasks to specialized helpra agents or plugins, all without you explicitly prompting it to do so. This rumor has led people to dub GBT 5 as potentially agent GPT, an AI smart enough to use other AIs. Now, OpenAI hasn't confirmed any built-in autonomous agent feature, but it's not outlandish given they've been working on an operator agent and plug-in ecosystem. It fits the pattern, unify everything into one model, including tool use. But if true, GPT5 might be able to decide on its own when to browse the web, run code, use a plug-in, etc. to get you the answer, rather than waiting for us to say, use tool X. that could make it feel a lot more like an AI assistant that acts autonomously within safe boundaries. In practical terms, you won't have to pick a mode for brainstorming versus coding versus vision. The same GBT5 can do it all seamlessly.

Another feature Sam is excited about is long-term memory in Chad GBT, which GBT5 will likely enhance even further. OpenAI recently rolled out a beta memory feature that lets Chad GBT remember facts about you from prior conversations. — Memory is probably my favorite recent Chachib feature. And now that the computer I feel like it kind of like knows a lot of context on me and if I ask it a question with only a small number of words, it knows enough about the rest of my life to be pretty confident in what I want it to do. He also notes some users may not like the AI remembering things, but most do find it helpful and you can always turn it off. GBT5 is expected to build on this idea of persistent personalized context, pushing toward a world where AI will just have unbelievable context on your life to give extremely tailored answers. In other words, GBT's context window and its ability to retain user specific info should expand greatly. In April, Chad GBT4 Plus users gained the ability to reference all prior chats, hinting at what's to come. With GBT, interacting with Cad GBT might feel more like talking to an assistant who actually knows you and remembers your past conversations in detail. One of the most viral rumors is that GBT 5 will have an insanely large context window, possibly on the order of 1 12 million tokens. For comparison, the current GBT4 maxes out at 128K tokens, about 100,000 words. A million tokens is like the length of five novels combined. Essentially, rumors suggest GBT5 could intake entire books or even multiple books worth of text in one go. This means you could paste a huge research paper or a lengthy legal contract or a whole codebase and GPT5 could analyze it all in one shot without forgetting earlier parts. Also, a well-known AI researcher, David Shapiro, speculated on X that context windows could be substantially larger than people expect, maybe up to 2 million tokens. And there are reports that Google's Gemini model is targeting 1 million plus tokens. So, folks assume Open AI will match or beat that. However, no official confirmation. Technically possible with the new supercomput, but we'll have to see. If true, though, it's a gamecher. pair that long-term memory with an autonomous agent that can take action based on what it remembers. And you're not just chatting with an AI anymore. You're working alongside it.

OpenAI's separate OE models have been all about stepbystep reasoning and problem solving. The kind of chain of thought approach that lets an AI work through complex tasks. Those models like 01, O2, 03 were specialized agents that could for example break down coding challenges. Now with GBT5, that capability is being merged into the main model. Sam Alman has hinted that the progress made with these reason models gives a preview of GBT5's power. How good are they? — Our very first reasoning model was like a top 1 millionth competitive programmer in the world. 03 is the 175th best program competitive programmer in the world. and maybe we'll hit number one by the end of this year. — In other words, by using clever training techniques like reinforcement learning for reason steps rather than just scaling up size open AI achieved with O series models, what we might have expected from an even bigger GBT6 size model.

model. So what happens when they take those reason and breakthroughs and bake them into a much larger general model like GBT 5? Sam is very optimistic and remember today's models already cut research cycles in half with smart prompting. Our team ran a few dozen sponsor email threads through Chad GBT. It mapped every winning negotiation path in 2 days. Work that used to take our marketer 2 weeks. And now our new negotiation strategy brings us two times more revenue from collaborations than before. We share these proven prompt and strategies in generative AI 101 inside AI master membership. Alman explained that the Oer approach doesn't improve everything uniformly. A reasoning tuned model might excel at coding or math but still be average in open-ended chat. However, if you pre-train a much bigger model and then apply these reasoning techniques, you get the best of both. With a significant jump in model size, compute and integrated reasoning, Alman said he expects the first bits or sort of signs of life of genuine new scientific knowledge to emerge. That is a bold claim. It implies GBT might not just use existing knowledge, but actually discover new science or prove new theorems on its own. This level of AI capability was previously imagined for GBT6 or beyond, but OpenAI now hopes to see early signs of it in GBT 5. To temper the hype, it doesn't mean GBT5 is an omnisient genius that will solve science overnight. But OpenAI thinks it could start tackling problems and generating insights that no AI has before. Perhaps we'll see AI assisted discoveries or proofs of concept in fields like math, biology, or chemistry coming from GBT5. A glimpse of AI as a

true research partner. GPT5 is confirmed to be fully multimodal, unifying text and other media. We won't get all the way here with GBT5, but eventually we do want one integrated model that can like reason hard when it needs to and generate like real-time video do that. — OpenAI hasn't explicitly said GBT 5 will handle big videos, but it will handle at least text, images, and audio in one model and possibly simple video or more complex modalities as well. The idea is you can provide any input, describe an image, speak a question, etc. and GBT5 will process it intelligently, then respond with text or speech as needed. This one model for everything approach removes the friction of switching contexts. For users, it should feel like more cohesive AI that can see, hear, speak, and think through problems in a unified way. Let's take a quick break from GBT5 to look at another cuttingedge video AI development. Imagine you could create a professional product demo video without any cameras, studios, or human models. Well, that's exactly what Top View Avatar 2 is doing, the sponsor of today's video. Avatar 2 is the latest AI tool from Top View AI, and it's basically one of the most powerful AI avatar tools for product showcasing. Picture shooting a promo for a sofa, barbell, or winter coat with zero cameras, studio lights, or human talent. That's easy magic. Upload one product photo, pick an avatar from a library of more than a thousand realistic faces and body types. Paste your script and seconds later you've got a scroll stopping UGC style video. Each avatar lip syncs flawlessly in over 20 languages, throws natural gestures, smiles on Q, and keeps every logo and texture on your product. Pixel perfect. One total control. Switch to manual mode and tweak angles or poses. Prefer zero friction? Hit avatar 2 mode and the AI handles everything from stage into subtitle styling. Brands love the math. Videos in under an hour. Instead of days, cost slashed compared with studio shoots or influencer fees and effortless localization. Generate Spanish, English, and Mandarin versions with a click. Shopify and Amazon sellers have already pumped out hundreds of Avatar 2 clips during beta, reporting higher click-through rates and faster AB testing because they can crank out content for an entire catalog overnight. Need 50 variants of the same ad. Regenerate with new avatars or scripts. Top view says from product image to full video in minutes. And that slogan isn't just hype. Click the link in the description and grab a free trial live now at Top View AI. So, if you are in marketing, e-commerce, or just an AI geek who wants to create a fake commercial on the fly, give it a spin and watch your products spring to life. No actors, no set, just pure generative creativity.

creativity. Training a model as advanced as GBT5 requires astronomical computing power and OpenAI has been preparing for this behind the scenes. Sam revealed a new internal project code named Stargate, which is essentially a massive AI supercomputer cluster built for GBT5. How massive? — I'm thrilled we get to do this in the United States of America. I think this will be the most important project of this era. — This aligns with OpenAI's historical trend. Each major GBT model used roughly 10x to 100x more compute power than its predecessor. GBT3, GBT4, and now GBT 5. each jumped by huge factors. In fact, GBT5 has been trained on an unprecedented scale, possibly the largest AI training run in history so far. OpenAI reportedly partnered with Microsoft to build out this supercomputing infrastructure. Tens of thousands of GPUs, if not more, and it appears ready to go. What does all this mean for us? Essentially, OpenAI isn't holding back on hardware. They've scaled up to supercomputer level might to give GPT5 the capacity it needs. How big is GBT5? This is a classic guessing game. With GPT3 around 175 billion parameters and GBT4 rumored near 1. 8 trillion, some expect GPT5 to be multi- trillions. However, OpenAI hasn't given any spec details and Sam Alman downplayed the importance of sheer size. The era of parameter scaling is over. The bottom line, GBT5's training compute is so high that we expect a qualitative leap in performance, not just a small incremental upgrade. And with Stargate, OpenAI has built the infrastructure to keep pushing further in the future.

Despite the hype, OpenAI is being very careful about GBT's release. Alman has repeatedly stressed that safety is the gating factor. The model has been undergoing extensive internal testing and red teaming where employees and outside experts stress test it to find flaws. GBT5 will not go live until these evaluations satisfy OpenAI's criteria. We expect a tiered release. For example, Chad GBT Plus users might get GBT5 at some standard setting and perhaps a higherp priced pro tier rumored 200 month could unlock GBT's full capabilities like larger context windows or faster outputs. Altman hinted at this kind of stratification back in early 2025 that free users would get a solid model, but subscribers would get an even more powerful version. All this GBT5 talk has you thinking, "Okay, I need to get serious about mastering AI. " But GBT Pro tier is too expensive. We've got you covered. Inside our AI master membership, we build generative AI 101, hands-on course that walks you from your very first prompt to prolevel tactics we actually use in the real world. No fire hose of theory, just by-size lessons that drop every week so you stay current without getting buried. Clear demos, quick wins, zero fluff. And because you're watching this in AI master, you can lock in a full year at 63% off right now. Hit the link below, jump in, and let's level up your AI game together.

Ever since Altman said summer, people have been speculating the exact drop date for GBT. Some optimists think late July 2025 could see a launch, perhaps timed with some OpenAI event or even a stealth update to Chad GBT. Others think August is more likely, giving them a bit more test time. Interestingly, Altman back in February used the phrase weeks, months, which had people hoping for maybe May or June. When that passed, there was chatter that maybe GBT5 had been delayed. But now, officially, GBT is very real and coming very soon. One thing's for sure, the moment OpenAI releases GBT5, it will dominate tech headlines and flood YouTube with GBT5 tests and reactions. The AI community and indeed many everyday users are waiting with baited breath. If the model lives up to even 70% of the hype, it's going to be very impressive. We could be just weeks away from new era of Chad GBT capabilities. So whether you are excited or anxious or a bit of both, stay tuned with our AMS membership. We'll definitely cover GBT5 in depth once it's out. put it through its paces and see how it changes the game. With the agent feature already live and GBT5 just weeks away, it feels like we're entering a completely new phase where Chad GBT doesn't just answer your questions, it takes tasks off your plate. Thank you guys for watching this video and see you in the next one.