Transcription
So everyone says that GBD5 is the best coding model on earth. Even Kurser, Verscell, big YouTubers everywhere and even Elon Musk who's definitely not a fan of OpenAI said that OpenAI will definitely crush Microsoft. So if everyone is switching, why am I not? And why is it not even close? So in this video, I will tell you exactly what's good about GPT5, what's bad about GPT5, and what I am going to use going forward.
Hey, what's up guys? My name is Rob and I've been a coder for over 20 years. But now I teach non-technical people how to build things with AI. And I use step-by-step video lessons and an exclusive community where I personally support you in building your first app. If you want to learn more, the link is in the description down below.
So I spend over 10 hours pushing GPT5 hard, mostly in cursor, and it's definitely OpenAI's best coding model today. But there are problems that to me kill it for everyday coding. And here is my brutally honest opinion. Let's start with the good because there are good things about this model.
Number one is the price. It is way cheaper than other top tier models. About 12% cheaper than Claude Oppus, 50% cheaper than Sonnet at the same price as Gemini, but better at coding in my opinion.
Number two, when using GPT5 in chat GPT, it auto adjusts the effort. So, it's super quick. It can answer like this if it has confidence in the answer and otherwise it will go away and say, "Hey, I need more time. and give me a second and then it will answer you and it's doing so very intelligently. Very smart."
Number three is simplicity. There's no more 03 GP5 40 GPT 4.1 or 4 or mini. You just have one model simplicity now. You can get chat GPT 5 or GPT 5 and that's it. You can then tell it to think more without swapping models, go on a search or really do anything else that you need it to do and it will just get it done.
And number four is creativity. It's definitely the best language feel of any open AAI model so far. It's better at writing. It's better at creativity. And it just gets you more, if that makes sense. It just feels more natural overall and less like an AI.
You know how typing long prompts into cursor or chat GPT takes forever? I've been testing this new tool called Willow Voice, and it helps me code way faster because instead of just typing to the AI, I just hold the function key and say the prompt out loud. find bugs that could lead to critical errors and propose simple solutions to fix them. Then I let go and Willow just types it out for me almost instantly without all the and awkward phrasing because Willow actually cleans it up on the fly. And it works everywhere, even in your emails, documents, Slack, and Notion. And it understands the context to get the tone and words correctly, too. You can think of Willow like the Apple's builtin dictation if it was actually smart, accurate, and fast. It's kind of like an AI for your voice. Plus, it is 50% more accurate than many other tools, writes almost instantly, learns from your personal style, and best of all, it's totally free to try. So, if you are a developer doing a lot of prompting, or you write a lot for work, download Willow Voice now. I've put a link in the description down below. It's a fantastic productivity tool that helps you get done more in less time.
But this is where the good things stop for me because there are two main reasons that make GPT5 unusable for me and I'm not joking in everyday coding and building apps with AI.
Number one, remember how I just a moment ago said that GPT5 is super fast? Well, in practice that isn't really true, especially when used in coding. It's quite the opposite then because in cursor you don't have a quick version of the model. You just have reasoning variants. And even if you use something called GPT5 low, which supposedly thinks very little, it can think for 5, 10, 20 seconds before doing really anything. And on bigger tasks, especially if you use the standard GPT5 or the GPT5 high variant, it can sit there for 15 to 20 minutes. And after the free week that we're all getting right now, you will not only pay for this with your time, but also pay with actual usage as in money. Because think about it, thinking cost tokens, and tokens cost money. And the best part is that meanwhile, Sonnet and OPOS have finished the work because GPD5 is still stuck thinking before it calls any tools.
And then the second thing is code quality because listen if if the output of GPT5 would be so good that it would be worth waiting for 15 to 20 minutes that be worth it because you could just run it, make yourself a coffee or even better run it in the cloud, get notified, come back and be like, "Wow, that was amazing." But that is not the case. Just this morning, I had a task run for 15 or 20 minutes and the output was unusable. It was completely bugged. I then had to switch to Sonnet, which supposedly is way less intelligent than GPT5, and it fixed it in just a couple of prompts. That's time, energy, and money wasted. And from the MS and replies that I got on X, lots of developers feel the same way. GPD5 encoding tools is overly cautious. It thinks too much. And my guess is that they left the reasoning very high even on the low variants to make it look very smart in demos. But what happens in reality is that just everything takes ages and it makes it unusable to use. If you really want to just build stuff, you know.
But hear me out, it's not all bad. There are parts of GPT5 that will be really useful. I will use it myself like I used 03. It's deep planning, tricky buck finding, and diving into nonobvious performance and security issues. It's a very, very strong thinking model.
But what am I doing now? Well, here's my AI stack. For everyday coding, I use Cursor with my Ultra Plan and Sonnet because, well, it's just the best editor out there. But for heavy lifting, I will continue to use Opus in Clawude Code because it gives me nearly unlimited coding. And let's be honest, if you try to use OPUS in Kursa, even the $200 a month ultra plan will not be enough. And then for heavy strategy sessions, so bucks planning, understanding things, I will likely use GPT5 inside of Kursa because it's cheap. And when I want to understand things, then the reasoning turns into a superpower just like it did with 03 or today Grock.
So my advice to you is this. Don't switch your AI stack today just because everyone tells you that you should. Give it one or two weeks the typical hype and die cycle and then evaluate if not only GPT5 is faster because maybe cursor and other tools will optimize it but also if people actually end up using it because I have a feeling that they won't. Then and only then make a decision because for now I am sticking with clawed code and I think you should too. And if you want to build apps with AI yourself, check out my AI coding blueprint in the description down below. I will help you build your first app, and I will not let go until you did. But until next time, I see you then. Peace.