GPT-5 vs Claude Code: Building an AI Video Generator

Okay. Sound looks good. Yeah, I think so. Okay. I saw a question here from uh Hello Edgar, you're up. Have you done some testing? So, uh, David, nice to see you here. Did GBD5 kill Claude? No, I don't think so. But they definitely made a leap forward in software dev space. I think 50 million tokens. Okay. I'm interested to hear your first impression. I guess the tool calling is going to be good, right? That was kind

of my late night tests. Tool calling it looks very good. So today I wanted to try to build exact same project on um claude code with uh son set and opus and I'm going to try out uh cursor agent beta might fail spectacularly but let's try it. Tool calling is ex exceptional instruction following as well but you have to relearn to speak to it. Yeah, GPT5 is great at tooling. Uh cursor augment pretty good job. Real

life test performs worse than GLM 4.5 says Florian. That's pretty crazy. So what do you run? Florian agentic coding with tool calling. Hello Ahmed GP5 and other tools almost take some jobs from front end devs. Yeah, we'll see. Or at least changes their jobs. Uh models before often like wild west winging it. G5 is much more cautious. Okay. Yeah. Hello Fruity. GBT5 loses context after a copper back and forward in chat. Hm. Is that in like

the So kind of the only thing I miss if you are kind of in the chat interface here. I miss my 03 selection. I really miss it. I went 03 did a search. Depends on what you run it on. And a lot of people say the model is bad, but the truth is uh the rapper they use is not great. Okay. Coding some stuff, some ideas, some creative solutions. Okay, Florian. I guess find what works for

you. That's the most important thing, right? I have a really time hard really hard time giving up cloud code. Uh, I'm going to try cursor CLI. Do I want to go back to cursor? We'll see. For me, I really enjoy claw code, so it's going to be hard to give it up. But let's see. 300k context at several points. Fall off wasn't too bad, but some fall off. Hello G. Uh G, nice to see you. Has

anyone given like the cursor agent beta any chance? I thought I could try it. Have you tested Opus 4.1 yet? Uh yeah, a bit. But I thought I can do it now just for at least I'm going to select opus to create a plan for this project. I don't think you need opus to do the project but we can plan it. I will stick with cloud code as well for now. Uh Anthony probably me too but

I'm going to give cursor agent just to see what it is. If Gemini 3 comes along with the Gemini CLI, I might try that out. There's augment code. There's there's a bunch of choices now we can try. I haven't tried Ko, but uh Cloud Code has like a special place for me, so we'll see. I'm going to go through the project. Uh I'm going to do uh it's going to be like a parallel project. I'm going

to do it in uh cursor agent hopefully with GD5 and I'm going to do it in cloud code with set and opus. Hello mcbab gt5. I'm 100% sure to stick with cloud code. Okay. I think we really need some time to get new system prompts for all the coding tools for GBT 5 to work well. Mostly open code per which perform not as well as cursor. Okay. I think it's nice to have options too, right? If

everyone was stuck in claw code, it would just drive the price up and it's already pretty expensive. So I think all competition uh is welcome for uh the our we that are consumers. Competition is great and it will drive prices down. If I have to reach for a CLI, it will be always cloud code max subscription. Yeah, CLI ID is good. I like that you're not locked into any ID and can jiggle things around. Yeah, and

like I said, competition is great for consumers, so yeah, I don't want to be stuck with only one choice. Gemini is behind the corner. I hope so, too. Cursor CLI has potential also like a max subscription. Yeah, let's try it out now. Uh I looked I don't think they have like a plan plan mode but uh I'm just going to instruct it to give a plan. So you could pick set and I'm not quite sure if

I going to run out of tokens. I haven't tried it yet. Twoacre studios. Nice to see you back. Christian hello. GPD5 is a cheating great option for people who can afford the max tiers. Yeah, I agree. Mbp. Yeah, I checked out Poly Market. We did it kind of live on stream yesterday. I kind of we kind of looked at uh That was the wrong one. We looked at um best models by end of August and before

during or right before the stream it was like 73 open AI 25 and during the stream it just flipped kind of watched it live. Uh, so now it's up to Google to follow up on this. Yeah, it was just um pretty interesting, but like I said, competition is good. We welcome that. So, I think I'm just going to start uh a bit on the project. So, I thought I could maybe do this into a video. So,

it's just going to be like me talking to the camera for a bit, but you can follow along if you want to. I I'm going to stop in between, but I thought maybe I'll turn this into a video. So, I'm going to kind of explain the project I'm going to build and kind of go through the setup and then I'm going to run it. So, that is my plan. $100 US a month for saving 10 to

20 hours a week. Yeah, that's worth it, right? For sure. But if you have like three or four $200 subscriptions, I don't think you have time to use them all. Just got to pick one, I guess, or two, maybe. to stretching it. So yeah, I think I'm just going to go through my project today. Which is better for backend claude or GT5? Maybe Edgar has a good idea about that. I haven't tested it enough. I don't

have like any indepth options. Maybe some other guys have some options. So yeah, I thought I'm just going to start now. I'm going to kind of check if I'm recording. Yes. And I'm going to do kind of go through the project. Hello capture. Okay. So today uh I wanted to try out both cursor agent. I want to kind of compare it to cloud code and I want it in uh cursor agent. I want to run GPT5.

Now there's like a low mower out here. I guess I have to wait a bit then. The microphone kind of picks up the lawn mower. It's very annoying. Working on a complex MCP right now which uh I want to provide to all of you. Okay, that'll be cool. that gunner. Sorry about the lawn mower in the background. It's very annoying. So, the sentiment kind of on GPT five, I think it's been mixed. You don't hear the

lawn mower. Maybe I can get away from it. I think sentiment wise people It's mixed. I think this did not turn out. Instruction following unmatched long running is exceptional, but it's still an alam. Yeah, I agree. Jeep. Yeah, those graphs on the stream are pretty crazy, right? Where are they? What is this graph here? It's very strange. Hello Bisco. Yeah, it might be time for Open AI to go to some pre-recorded stuff, too. They're not doing

themselves a favor with this live streams. So, I think some of the I think during the hallucination part of the stream, the model did hallucinate something. What's the point? Why just do some pre-recordings and just Yeah, I don't know. For the seems very strange. Try augment. Quite impress impressive and works well on larger code bases. Okay. So, I don't think I have time to wait for that lawn mower to finish. I'm just going to see. So

annoying. Okay, I was life GP5 is good at some things, but it seems uh to almost as bad alignment safeguard training as the OSS models makes it kind of crap at communication chat applications. Yeah, I guess chat is kind of left in the dust, isn't it? Okay. So you here okay I'm going to do the I don't think uh there is any other model that is as good in broad tool calling and context performance and instruction

following that is good for agents. So for me I'm pretty happy but people have other use cases right? So maybe in like science you might want to consider maybe doing Google Gemini deep think. We'll see vibe coding. Some people do vibe coding. So there are different use cases. Uh today is going to be yeah almost like a small vibe almost like a vi a a aentic project but yeah you can call it v going to so

I'm just going to go through my project now because uh just wanted I might turn it into a video. Okay, so today uh I kind of wanted to look at the new cursor agent. Uh it's still in beta, but here we can kind of pick the GPT5 model from OpenAI and I want to compare it to Claude code and I'm going to do like a combination of set and opus. Right? So I'm going to do opus

for planning and set for execution. Uh for cursor it's just going to be a GBT file for everything. So to trying to compare them um so to compare them uh I wanted to do like a not an advanced project but it's a front and backend project with some uh yeah workflows or pipelines. So basically the core concept is going to be to build a web app that generates shorts videos from a test test text. I'm going

to do that again. So basically I'm going to build a web application that generates short videos from text description using AI with a clean and modern interface. And on the backend side we're going to use um OpenAI GPD5 to plan out the prompts. We're going to have an AI agent or let's call it a tool call with the web search feature from the OpenAI API to gather some context for any topic the user enters. Uh then

we're going to create some scene planning and image generation using image genen4 and we're going to aim for two video clips merged together using ffmpeg. Uh the video model is going to be the video model is going to be van 2.2 uh kind of hosted here on fal ai. So it's not going to be v3 or anything. So let's see how that works. Just wanted to try it out. So, it's pretty straightforward. And yeah, like I

said, the idea is just going to generate some video prompts and we're going to turn that into two clips and merge them with ffmpeg. We need a UI. So, uh we kind of want a front end UI, right? Uh for the technical stack I kind of have a set UI here because uh it's going to be react based tailwind CSS and we have some instructions here right uh of course uh GBD5 needs to build like a

backend front end uh bridge here or like a pipeline to solve this. Uh I have stated where my documentation is located and my uh environment variables here. Uh we have some text stack selection. Uh yeah that's pretty straightforward and some output specifications. And for the documentation I'm going to feed it. We have GD5 uh just a simple how to use GT5 here uh in JavaScript. basically a shot completion image gen 4 docs open AI vision if

you want to use that I think I'm skipping that we have the open AI web search tooling how to use that web search tool example in JavaScript and here is the video model documentation so this is what we're going to give uh for documentation for the project so I kind of preloaded that and here is uh the prompt uh I'm going to feed it. So the last statement is going to be generate a detailed specdriven development

plan for the app. Right? So that is the idea and after we get this plan uh we're going to move on to the aentic coding part and actually build the app. Okay, I need a break. Hello Unreal. I have not gotten the specific GPT chat model to do reasoning. I don't think it has reasoning in Chaos Life. If you go to the playground and you select is it GPT5 chat uh it doesn't have any reasoning here

you can see if you do five we can select effort here reasoning effort search and output length, but with the GP chat, I think it's just created to be fast and responsive. So, there's probably not going to be any reasoning on that model. Yeah, I think so. Someone said minia was pretty good, too. I might give it a shot. So that's probably going to be over the weekend. Nano, haven't tried it. But for the project, it's

going to be [ __ ] hopefully. Yeah, I think chat is more for not for doing other thing than chat thing. Okay. So the way I'm going to do this now is on claude I'm going to go into plan mode. So I'm going to do shift tab tab plan mode on. Right. And I'm just going to say something like you go. I hate this. Why did they do this? So, I'm just going to say something like read and

execute uh this prompt domarkdown file, right? Uh so this is going to go into plan mode and we will get our plan uh from the prompt we just went through. So I'm going to start off cloud code on that. Uh did I select opus now? I think I forgot that. So I'm going to go and select opus for the planning part there. And I'm going to do read and execute uh prompt MD. So in planning mode

here, it's going to look at the prompt and we're going to start to generating this SD plan for our application. So yeah, this is uh firing up uh on cursor agent, we don't have that plan mode. So I'm going to do it manually here. So read prompt MD create a plan for the project SD okay specdriven development. So that is going to be the cursor agent using GPT5. Okay. So the idea is that we can kind

of run the exact same setup both on cloud code and on cursor agent. So let's watch here now. So you can see GPD5 here now is using its tool call looking up uh prompt MD. It found it. It's reading it. It's reading it. So it's pretty similar to cloud code um using its tool call here like in combination with the cursor uh CLI here. Hello Vid Rick. So we can see on cloud code now that it

already kind of has uh gone through a plan here. So, I'm going to take a quick look. Here is Claude's plan, the video generation app, specdriven development. So, we get this really nice structure here with all the frameworks we're going to use, the API calls, the video processing. We get our file structure. This looks nice. We get our faces, research, uh, image generation, video generation, main generation. So, this has some API endpoints we can call. We

have the front- end development in one section. So, we have some API integration details. So, we're going to use some JSON here, I think. Uh, we have our environment environment variables. These are hidden. Okay. So, let's see. So, I can do yes and auto accept edits. Okay. Uh, but I'm going to stop it here because I'm going to swap the model back to set. I want to run set on the execution. Continue. Right. Okay. So, if

you go back to GBD5, it's kind of still generating the plan here. I think it's a bit slower GP5 after my testing yesterday. I think so. But, uh, doesn't matter too much. Just going to give it both the time uh it needs here. So I guess cloud code will get like one point more if it's much faster. So now from the plan we kind of have our huge to-do list here that set is going to execute.

We have all the implementations. We're starting to do some work here. Right. So yeah, looks pretty good. So, you can't really see it here because I'm in the wrong folder. But, uh, if you look here, we are starting to generate this plan here. Okay. created a comprehensive spectory event development uh based on the prompt nd and saved it to docs video generation app s uh dtmd. Okay, we can take a look at it. So this is

the specdriven development plan for uh from GT5. We have an overview. We have the targets, key requirements. We have the allocation of the documentation. Yeah, should be good. We have an highlevel architecture, detailed component design. We have some roots here, pipeline. Okay, looks pretty solid. Very detailed. We have some file system layouts, ffmp commands, ffmpeg commands. Okay, so very detailed plan here from uh GID5. I'm not going to go fully through this, but I might share

this if people are interested. Okay, so I'm happy with this. This is kind of going to be the plan we're going to go through. So based on the plan uh create the app. Okay. So now we are getting started on GT52. You can see cloud code is already getting started here. So I'm just going to give access. So, I'm just going to let this run for a while now and we're going to keep an eye on

them. Okay, just need a little break. Uh, but uh, yeah, [ __ ] is a bit slower, but uh, the plan looked really good. Hello, Talas. Uh, no, you're not late, Graham. Nice to see you. Yeah, one thing about GPD5 is the pricing is very good, right? Uh I'm on the max 100 on the cloud code. Unreal. Any good tools to generate wireframes? Not quite sure, Muhammad. I don't see scaffolding the next JS front end fast API back

end and basic pipeline. Okay, so claw code is already a good chunk here on the way. Quick re recap. What is your view on GP5? Uh, is it time to switch? That is what I'm testing out today. Killing joke. I am um trying to build the exact same project both on cloud code and on uh cursor agent with GPT5. So this is kind of the first time I try it. Okay. So it's going to you can

see here GP5 is going to make some directories. So I'm just going to I'm going to add shell uh make directory to allow list. So we created some directories here. Do we have to pay for GD5 in cursor CLI? I'm not quite sure. I tried to put in my API key, but uh I don't know what happened. So, I'm just going to run it until it stops uh killing uh Graham. Seems like they have a free

option. You can try it out at least. Clo is now generating our FFM commands for video merging. It's doing pretty It's pretty fast though, at least compared to this. Diver, welcome as a member. Really appreciate the support, man. Thanks for dropping by. Really appreciate the support. Okay. So, cursor agent asked me if I can uh write to env. Yeah, I'm going to put it on the allow list. I like that they are not displaying uh my

API keys here. I appreciate that for the stream. Both agents now is working hard. We created a readme. GPT5 is free on cursor and Windsurf right now. Yeah, that's cool. So, we can try it out. Perfect. Oh, we have lots of time to test it then. I'm going to test it a lot here on the cursor agent over the weekend. impressions so far. Killing joke I think um very good at tool calling um instruction following seems

good too and a lot of other people have the same opinion. What do you have you given jdify about test usage limit? There might be some usage limit Sanji I don't know but uh let's just try it. Okay. Okay, so if you go back to cloud code now, you can see it's already quite far down the to-do list here. We are on actually building out the the front end here now. Uh while GPT5 is kind of

still working a bit on the back end here, I think so for speed so far uh we started of course first on cloud code. Uh but I think cloud code is a bit ahead on speed but uh doesn't matter too much in every case. So, let's see what happens here. David, I did hit I did not hit any limits by running the normal GP5. Okay, nice to know. Claude is the best, says 24 AI news. Hello,

Juan. Nice to see you back here. Have you been trying out GPD 5 so far? Looks like uh GT5 on cursor agent here is testing out some uh some servers here. Hello Akbar. Uh yeah, it doesn't matter too much. You're working 24/7 and I'm using together with cloud code and GT52. Cool. What's your first impression? One. future. That is a very good guess because yeah, I'm from Norway. Nice pickup. Your impression is that you love Antropic.

Okay, that says a lot. Now, let me create the final setup file and test the application. So, pretty high speed here by cloud code. It is running the set model. So, it's a bit quicker than Opus. I guess is this stuck on this uh mpm rundev here. That's not good. I might have to stop it. We'll see. Hello, Shane. I'm kind of in the middle of the testing now, but so far good on tool calls and

instruction following. Do you notice any difference in speed with Opus 4.1 and Opus 4? Haven't tried it enough yet, Tom. I think to give a good answer. Yeah, this looks kind of stuck on I'm just going to Can I even stop it? Okay. Ctrl C. Uh continue. Okay. So, we're going to do the build here on CL code. We got some probably just some small root errors here. Yeah, API root errors. Uh, we kind of got

stuck on the rundev here in um, so we didn't have a timeout on that. It's testing out the server with a prompt. A calm ocean sunrise. Okay, that's good. Sleep too. Did we get our response? Windsurf gets stuck all the time on agent mode. Yeah, I know. Night spider. Is this some TypeScript error or TS errors here? Okay. So, kind of looks like kind of looks like GP5 is finished. We didn't kind of see all the

instructions. So, how to run? Create a v environment and do a pip install. I can follow these instructions. That's fine. Could have been nice if they did it, but uh good instructions here, but um let's make it simpler. Hello Nion near Nero John. Did you figure out the PayPal thing? Daniel, it's usually on npm rundev that it gets stucks. Yeah, I noticed that too. So you should run npm run silent grap ready. Yeah, maybe something that's

to have in a claw md file Daniel or something like that. Sometimes there's like a time out too. So time out 10 seconds or thing. Hello, Louis. Jack, what's your conclusion on GBD5? Should be canceling clo max and why? Hello, MS. Uh, Jack, maybe it's a bit earlier yet. Um, I'm kind of testing it now on the new cursor cursor agent CLI. Uh, so far we think it's pretty good at tool calling or very good and

instruction following seems pretty good but uh bit too early. I think I'm just going to wait a bit for CL code here. Well, let's try it, I guess. Can you summarize what you're building uh with cursor CLI? I can do like a quick summary like it's basically you want to see it now anyway the idea was to create a simple web application that generates a short video uh we have some pipelines with some research and we're

going to create some uh AI images some video and we have a UI and a back end is it going to work let's Okay. We do have a one UI at last. So I'm going to stop that and I'm going to say the same. I told Okay. So I'm going to do a part for the video here. Okay. So it looks like both uh the agents now or the CLIs have done their part. So I said

to GP5 here an intern needs to run the app. make a simple instruction to follow to start the servers and start the UI. So I'm also going to give that instruction to cloud code right. So we are on kind of the same playing field and let's try to follow the instruction it gave me here. So it says uh the first thing we need to do uh let's see uh brew install ffmpeg uh I already have done

that um prerequisites but we can try it. Okay. Uh, we should have this. I know I have this. So, I'm going to skip this part. So, I'm going to start here on one, right on the fast API. So, we need to go here to services. So, we probably need two terminals here. So, let's start with the back end. We're going to go to services. We're going to start uh create an environment here. Okay. Uh, we're going

to activate it. Okay. Uh, we're going to do pip install e right? So far so good. We're going to run this command here to put it in the path. Uh, it's already in use. Okay. Yeah, I might. So, let's say the intern ran into this issue. Okay, so we have an issue here. So, I'm going to copy it. Uh, I'm going to copy the error in here. Hello, squad. Phil, nice to see you. Uh, I got

this error. So, what's the opinion on GTA 5 versus Opus 4.1? a bit soon to tell at squad I think. I hate this. What is happening here? You have already something on 8081. Pick one. Uh, we can kill the part. Okay, we can kill it. So, let's find the P uh P. So, let's find this P. It's 418. What did I say? It's Can find it here. 4181. Okay, we killed that. Now, let's try to run

the server. Okay, perfect. So, we have the server running. Good. Now, we can go here and we continue. Let's go up again. So, we're going to start up our front end. We're going to do CD web. Perfect. We're going to do an MPM install. We're going to do a generator URL localhost 8081 MP and rundev. Okay. 8081. We have another issue here. So here 30,000 is in use. So, let's see what feedback we can get on

this. It's fine. So far, it looks pretty good. But this is uh Hello SQL. Yeah, I'm I'm just going to follow the instructions. I'm not going to interfere here. So again we can try to kill port 3000 4708 40 708 Okay, we can do it again. Okay, perfect. So, let's take a look. That was quite boring. Okay, so we have it here. Now, this is from GT5, an AI video generator. So, we're not going to try

it out yet, I think. So, this is from cursor agent and GPT5. Let's just try it. Okay, so let's see if this works now. So uh we can enter a detailed description of the video you want generated. So I'm going to think of something uh we have the option we should have the option to do some web search. So create a meme of the sentiment of the GPT 5 launch on August 7 2025. To be honest,

I have no don't think this is going to work. So the generate button doesn't work. It's very boring. So I'm going to instruct the terminal here. Uh the generator button is uncclickable. Is the back in a mock? I don't know. Let's see. Let's check the code here. And it has a code here, right? It has some backend code. It should be it should be up and running. Yeah. Fast API. Okay. So, I'm going to say to

the generator button is uncclickable. Hello, Jake. Close source. Tom Pelican riding a bike. We did that yesterday, but we can try it later. Was Horizon better much better than GD5? I don't know. Sergio, have you heard who is the provider? Yeah, okay. So, we tried to fix this. So, let's see if we can do um refresh. So, let's try now. Okay. Did this trigger something here? Okay, so we did trigger something basically. It's really hard. Doesn't

look like anything is happening. Okay. So, I'm going to do I clicked uh server responded with servers. So, we're going to compare this to Claude Codes version of the app later. We had a quick look at the UI and I think it looked a bit more impressive. But, of course, cursor agent is of course in uh in beta mode. BMAD is great, but you can't create your own PRD and making it break into sprints without any

frameworks. That's interesting, squad. A lot of people have been talking to me about BMAD. Uh, I think it's time to try it out next week. About time. So, let's restart the servers here. We can just restart this. Okay, we tried to make some more changes. Let's try create a meme of the sentiment of the GT5 launch. uh yesterday. Are you kidding me? Oh my god. I agree there with us, but I kind of wanted to try

to test cursor CLI and GT5 in one time. So, I'm kind of used to CL code, so it's a bit frustrating, but let's give it a chance here. Still early results, but what do you think, Ken? Did you Did you try your PHP from yesterday? BMAD works better in GIF 5. Claude is better writing complex code. Unfortunately, I was hoping for better. Disable. When did it do that? Why? Okay, we tried to run some debugging. Let's

try it again. Let's look at the server. Okay, we have a a get here now. But uh let's see. So this is an event here. Researching, planning scenes, generating images, generating videos. going to have to in ask to investigate, but I think there's a few stuff uh gone arai here, but usually, yeah, the console log looks fine, but Uh um jobs ID events we have the ID and I know nothing about front end. So I got

to give it a few more chances. Let's switch to cloud code. If cloud code has the same issues, maybe it's my planning that's bad, right? So, I'm going to give it one more chance. Start the server again, and I'm just going to let it run. Okay. So, while having some issues here, uh, uh, it says API jobs post API generate. So, let's just let it run here now. Let's see what happens. Maybe it's going to use

a long time on the research. So, we don't know yet. So, this is just going to keep running. And let's go back to our cloud code app here, right? So, this should be pretty ready, too. We have our instructions. Uh, I'm just going to update uh port 3000 is in use. 8081 is in use. So we don't run into issues with that. So we should be able to run them con uh at the same time. So

we can switch to 3001. Uh yeah, why not just switch? I just told it to do that. Okay, so here is the instructions for the intern. So, we're going to navigate to CC test. Right. And the instructions is npm install. I can do that. It's fine. Uh, copy ENV example to ENV local. I'm just going to check the keys here for stream. Just need to fix my API keys here. It's a bit annoying. AI. Why? Okay,

so we did that. We added open key. Now we can just do an MPM rundev. Let's check out 3001. And okay, so this UI looks a bit more interesting, right? Create videos with AI magic powered by GD5 and Imagent 4. How it works? So, research and planning, image generation, video creation, merging. We have some powerful features. Start creating now. Okay. So, I'm going to do the same prompt. So, let's see. Research and planning. Okay. I'll be

surprised if this is a oneshot. Doubt it. Okay. So, we did send uh the query. Well, let's see. The UI looks a bit better, though. Opus versus Gypty 5 is not really a fair comparison. Besides, the cursor implementation may not be the best. Yeah, I know. I don't even know how long I haven't tried GT5 on this uh workflows before. I kind of look stuck to be honest. So we might have to switch that. So this

is the route it's using. Uh, I'm just going to open AI. Description must be a string. Open a key description. uh open. The task is to research the user subscription using web search. So GPT5 web search preview research and plan a 10-second video based on the description. Okay, it looks pretty good, but uh I don't really know what um the research tool is doing here. Research and plan video. Is that in here? Yeah. Research and plan

video. Okay. It's an async function. Okay. So, we did move on. Cool. So let's see now. Research completed on August 7. Uh open AI launch DP5 across GPT and the API blah blah blah. Scene count 2 video prompt two. We are generating images. Good. Image generated. Success. Okay, pretty good. Hello Benjamin. We are generating videos scene one. Okay, this is looking pretty good. Remember this is a one shot. I guess the app did an update. I

saw the release of GPT5 today. How is the testing doing? Uh we are giving it a chance now. We are struggling a bit with uh I'm trying out the new cursor agent. but it doesn't seem to run. So, I'm going to try some more debugging. It's stuck on Hello Shereik. It was fun. A lot of people here. So I'm going to do a bit from the bit for the video here. Okay, so we can now see

that the clawed version of the app has uh completed phase one. It did the research. Uh it found some contextual info on yesterday's uh API launch, yesterday's open AIGD5 launch. Uh it's going to do two scenes, two videos, and it kind of jumps into phase two. That's going to be the image generation for the video. Uh it's going to prompt it with uh some uh yeah uh two prompts here. And it's going to do phase three.

That is going to be the video generation. And it's going to do a motion prompt, right? And it just keeps on working. So I'm just going to let this run. Videos often take uh like a two three minutes to generate. So we're just going to let this run. But for claude now it's looking pretty good. I guess the UI here doesn't update in real time. But uh let's see if we have when we complete this. Is

it possible to have cl code use an open AI open MCP with GT5? Yeah, you can do that Graham. It's super easy to generate open GP5 MCP. While over on uh the GP5 version of the app, we have some issues. We kind of get stuck on this uh API request here to do the job. So now we kind of instructed cursor agent here to kind of check up on the back end, do some more testing and

see if it finds any issues. So yeah, it's going to try to do some more testing here. Uh let's try it when this has uh completed. Yatas, are you comparing GD5 and Claude Opus or Claude Sonet? So uh I did GPD5 uh in cursor agent and I did opus for the planning of the app and I did um set for the execution. I'm really hoping uh GT5 is entropic on nudge. Yeah, we'll see. Okay, so still

early days, but GT5 uh has already some excellent results in debugging your PHP codebase. Okay, so you're pretty happy, Ken, with the performance. Okay. So, you can see uh we made some changes. We updated some backend stuff here. So, I'm going to restart the the GT5 version of the app here. So, it did actually kill it. That's fine. So, let's restart the server. Now, it's already in use. Okay. So let's do this P ID 58. Call

it 58963. Okay. So let's run that. Restart this. So, the UI from GD4 5 was a bit boring, but let's try it out again. So, let's see if we can get this rolling. It took some time in to actually do this. So, we're going to give it some time. Also, this video generation looks pretty slow. Hello, Super Llama. What about Codeex CLI? They have GPD5. Yeah. Uh, I thought I can just do two things at once

today by testing out the cursor agent too. Work with GP5 and cursor all day long yesterday. It's great but not different from 4.1 at least for my use case. Yeah, that was kind of what I was thinking too. me personally don't have any like in extreme use cases like um insane big code base or anything. Eric says GBT5 has been in inconsistence and quite bad to be honest. Okay. What did you use it for? Uh, Eric.

Hello Thomas. Uh, I read something along the lines of chat GPT chat GPT5 isn't a model. It's like a system who decides where to route the request. Uh, I think it's false that it's not a model. That's not true. I think kind of more what happened is that um if you kind of look at the chat GPT interface now we have GD5 and thinking that's it right uh so what I think they did is that they

uh they kind of where is it they are kind of rooting between Jeep 55 nano and mini and maybe chat based on your input. So let's say if I'm on chat GPT and I say hello, this request shouldn't do a thinking model, right? It should be an instant response. We don't really need reasoning to answer this request. It's probably a mixture of experts composed with different models with expertise. Yeah, like a bit. Uh just checking back

in. How is your testing so far? Uh it's been mixed. Uh I think it did a good job on the tool calling, but uh it's kind of hard to say. It's a bit too soon. The router knows if you need reasoning or not. Yeah, that's kind of the idea, right? I tried it with simple 300 lines of uh point script indicator for uh TWW and it failed every time. Okay. Okay. So, we did actually move on.

Uh we created a video with van 2 v 2.2. Now we are moving into video two out of two. We put in our image, we put in our hard um we put in our motion prompt and now we're going to generate the video. So after this final video is generated, the idea with the app is to merge them and show them to the user. So hopefully we get an update here in the app with kind of

the final video generation. So remember this was a oneshot app on claw code. Okay. So we failed. So we failed to merge. Interesting. So that is something that has happened before. So we generated the both videos successfully but the merging failed. This should be an easy fix. So if we grab this, we just need a correct correct ffm command. So I can go back to claw code. I can say uh everything worked well until the merge.

Paste in the error. Uh find a fix you know how to initiate the best model I was thinking I kind of wanted to see if we do let's say we are on chat BT now and I do what is a kind of a simple thing I used to search. Let's do something stupid. Best beer in London. And I thought if I did think hard, is that going to trigger anything? Best beer in London. Tink hard. So

it looks like at least triggers something if you do this command tink hard. So now it's searching. At least we can also get quick answer. I've been coding uh a form plugin in the new CMS and my PHP codebase has grown to almost 3,000 lines including li libraries. Okay, that's interesting, Ken. That's pretty cool. Okay, so we made some changes to our um to our uh ffmpeg commands here to see if we can merge this. I'm

going to try a sneaky I guess we can't do that. That's a bit annoying. So, we don't we have a try again button, but that doesn't do anything, does it? So, we need to restart our server anyway probably. So, I can do that. So the UI is fine. Let's generate something. So let's do something a bit more simple. Let's do uh let's do something people's reaction to the GPT5 launch on August 8th 7th 2025. Ultra cinematic

shots. So, let's try again. So, I'm going to give it another shot. See if we can fix the merging. Okay, we are sending this. Good. Yeah, this is going nowhere. Um, doesn't look good here now for [ __ ] So, I'm just going to stop the servers and I'm I'm going to do this. I'm going to say there are some big issues. Uh we can't move past Uh re this sign this think hard. I saw a person from

OpenAI on X saying that think hard is supposed to work. Yeah, I think it looks like it's working. I'm going to try it here now. So I'm going to do try to solve this problem. Think hard. Can we see the 5second clips? Uh, yeah, we probably can. Or did I restart it? Let's see if I can't remember what I did. Okay, so I guess I'm remove them. Uh, but I can probably find it. Uh, let's look

afterwards. I can look it up after SQL. So let's see if we triggered any thinking here now. Switching the UI to connect directly to the generator service. Enabling course at the back end. So it's doing some more changes. That's it. I redesigned the flow to eliminate a flaky proxy. How to run this new setup. Okay. Let me try. 651 98 65198 Okay. Um, stop this. 3002. Okay, let's give it another shot. Okay, that looks better. Why

is it creating images? final video. Is this just mock or what is it? I don't I don't get this. This looks horrible. I don't know what's happening here. think ultra hard is better. Is it? What is kind of the difference? I don't kind of get that. There must be maybe some thinking budget right? Maybe it's something like Let's say where is everything? Let's say like think is like 4,000 budget. Think hard is like 10k and like

think ultra hard is 30,000 tokens. You think it's something like this? How do you configure MCPS on cursor CLI? I haven't tried it. Uh Ricky, not yet. In my experience, cursor with set is better than cursor with GT5. Gift 5 is very good. Uh not wipe approach. Yeah, wins. Uh it's a bit early. But maybe cursor also need to adjust to a new model. I did some testing today. See Opus 4.1 is just better says 921.

Interesting. What I don't get is uh how do we activate writer now Eric? Um, let's see now. So, let's check out the API here. So yeah looks like 4.5 is gone. Yeah, 4.5 is removed. So that means that we can't use 4.5 anymore. We still have 03 here, but 4.5 is gone, I think. Can't find it. Can't see 4.5 here. So one minute silence for 4.5. I didn't use it much, but I tried it out for

some brighting. Oh, wait. That was way too fast. That's one video. Why was it so fast this time? Okay. So, okay. So, I'm going to do this for the video. Okay. So, now you can see uh the clawed code version of the app or opus is finished. So, we get our research summary here. Uh basically we just have a bunch of resources. We have some information about GD5. Uh this is our generated scenes. So we can

see our two images here. We can uh click to enlarge. Okay, not bad. We can download them. And here is the individual clips it generated. Can we play those? Yes, we can play those simultaneously. Yeah, that's pretty good. And finally, uh, we were supposed to have the final video, uh, but this couldn't be loaded. So, let's see what the issue is here and see if we can fix it. But we failed to load the final merged

video in uh UI. Okay so this app uh is pretty good. It was pretty easy to set up. Is this even legal? What? This or what are you doing? What are you working on? instead of export uh load API key from let's the keys. Since the GPD5 roll out, legacy models are only available on pro team and or enterprise. Yeah. So on free and plus can you can't use them. Yeah. Okay. If it stays on $20

for a month, cloud code is still the best option. Yeah. Okay. So, we on the video model, we can turn off the safety checker. So, uh, that could produce like, um, NF, nonsafe for work video, I think, or like, uh, not maybe not safe for work, but like more risky videos. Hello, all gold. Dan, uh, we're just testing out building some, yeah, pretty straightforward video generation app. Uh, co Dab, nice to see you. Hello. So what

we can do now is um um uh intensive high action street fight. can't swipe in Rome. I don't know how what the hell is going to happen. Not much, Edgar. Uh, I think I'm just going to give up on this um this part. We'll see. So basically the clawed code version of the app worked kind of one shot uh from the GP5 version. Um just having so much issues with the same prompt uh that I'm probably

not just going to bother finishing it. Uh, this is probably going to be the last attempt. So, the images is just Yeah. So, I'm going to do a conclusion, but it's not really fair, right? Because I think we also tried a new cursor agent, so maybe it was not optimal. But since Cloud Code did it like one shot, it's kind of hard to ignore that. Now take the same prompt. Choose the best model on cursor ID

and see how it goes. We could try it in cursor. We do have uh I don't have any cursor probe or anything but we do have the we should have uh let's switch models. We do have GFD5 so we could try it. From my testing, large coding challenges combined with think hard um appear to trigger GPD4's most advanced reasoning mode even on the plus plan. Okay. Yeah, we can try to run the same prompt here in

cursor after this. So, I'm just going to see what we end up with here. Um, I think we might need to scrap this. I'm just going to wait for this for finish. So, yeah, think I'm going to scrap this. Doesn't seem to work. So, let's try it again from uh the cursor CLI. I think just that's just the way to go. Uh, Sonia, I have it. Uh, try dip in the ID. It's pretty good. Okay. So,

I'm on an GP5. Okay. I'm going to do the prompt. What was my instruction here? Can't remember. Close this read prompt domarkdown create a plan for the project uh specd driven development. Okay. So I'm just going to do this part for the video. Okay. So we gave cursor agent a shot here on GD5. Uh we didn't have much luck. So some people suggested we can just try it in the cursor IDE kind of setting up with

the agent and selecting GT5 here as the model right. Uh so let's try that. Uh so I'm just going to give the same prompt. We're going to do read the prompt MD create a plan for the project uh sddd. Uh yeah we don't need to send to background. We can just run it here. So yeah, let's see how this goes. Now we're kind of running it inside the ID here on cursor. So we are still we

are generating the first video on the other app on the cloud code version. Uh Edgar says um it's not without issues for me as well becomes down different interpretation of prompts. The models feel react differently. Yeah, we need more testing, right? I always felt uh I have an intuitive idea what works. Uh that's good and what doesn't. Uh this is kind of gone at the moment. Might also be due to new alignment. Yeah. And of course

um some I guess cursor had had a while now with GPT5 I guess to prepare for the launch. Uh okay so I read the prompt. I'm now generating a std plan. Right. So that's good. Uh we are going to create some API contracts here, responses, data models, some components. Going to have some prompting. Okay. So let's finish up this plan. So that was done. think um execute and build the app according to plan. I'll be interested

to see how your results compare to the same prompt but with think hard added. Okay, let's see how it goes without I might follow up on that. It is thinking here. So, it's triggering some kind of thinking or reasoning, right? So, I kind of like this more already. I liked when uh I like when they do the front end in like a modular and they do back end in like another modular cont. What did I miss?

Lol. Uh, you didn't miss too much. I'm just trying out to see what we can do with uh [ __ ] and kind of compare it a bit with uh with um Claude Code and Opus and Sonnet. Uh and we tried the new cursor agent wasn't that successful, but it's a beta version. So, we're going to switch tried it out on cursor IDE. Now selecting chip D5 and the agent. I'm just going to do a part for the

video. Okay. So the AI video generated from the claw code and set opus version. Now we did a input a high action street fight in Rome. Uh we have completed this. So we can see the research. Uh it's not a lot of research on this. It's just looking at kind of how the the Rome uh city is kind of looking. So we generated some scenes. We created this boxing scene here. That is clip scene one. We

have the other scene, two people fighting. Uh we have the individual clips and we have the generated video. So let's watch this now from the start. Uh, so you can see the first clip is going to be two people kind of fighting. I don't know what the camera was there. Uh, here are some other people. So, remember this is not the best this is not the best um video generated. I just wanted to try that open-source

video generator. Uh, but the app all in all after some small adjustments looks pretty good. Uh, we got the video, we got the images. Uh so all in all I think this was a very good job by claude code and not a lot we have to iterate on to go from uh input prompt to some research some scenes and some merged video with uh ffmpeg. So yeah, pretty successful app uh if you ask me here from

cloud code. Uh David asks this is the opensource I call it open source. It's like the van 2.24b. I just wanted to try it out. Uh it's the image to video version. It might actually be the 20 the image to video version. So, it's not the best video model, but uh I just wanted to try it out. You should have probably pick like foul or VO3 or something, but just for this test. It's quite cheap though.

And you can set a ticker called safety check off. So, if you want to do like more challenging content, right? Okay. So, let's check back on cursor here. So, uh, I really kind of like the what they did with the tool calling here for GD5. You can kind of see that we if expand this a bit, can we? No, you can kind of see that we are we can kind of check the tools it selected here.

So, read the page, read the package, and I think this is pretty good. So, it's still kind of working on our app here. So, while we wait for that, uh I think I'm going to try to add um maybe add a feature to our other app. So, I just saw that Sunno, not Suno, 11 Labs, uh launched some kind of music generator. So, I want to see if that is available in the API. So, it's music.

Just want to check if this is available in the API yet. Hello Nick. GT5 PL uh plus cursor. Uh I just started it. We tried the cursor agent. Uh we kind of failed on that but uh now we are switching to switching to the IDE. So let's see how that goes. Okay. Made by um made by Nando. So the API is not the music is not in API. Okay. So then I can guess I can just

skip that. Uh what other feature should we add then? What's this productions? Okay. Yeah, I can't see it in API. That's a bit boring. So, I don't think it has an API. I have used uh Sonato that has an API but I don't think Sunno either has an API start and end frame. Uh then I probably have to switch video models. I don't think this has start and end frame does it? We can try to switch

video model though just to see the difference. That's a bit boring. Yeah, that's a bit boring. Let's ask cloud code then what feature it wants to do based on the full full project brainstorm. The next impressive feature to add ultra think this is going to be set though. Uh there is a page in the documentation for music but it has coming soon message. Okay, pre-made templates. Make a pre-made template. Okay, so let's see now what uh

Claude Code suggest for the app for a next impressive feature. So we're going to do the ultra think keyword LIA by Google seems to be available in API. Okay, that's pretty interesting. Okay, so we are triggering the think here. It's going to say the background music and voice feature stands out. So let's check back on cursor here. So let's see it. Uh we spawn up a base app uh per this. Okay. Lydia Works quickly tested it.

Okay. Hello Said. Can I get one? We're going to do at the end of the stream. I thought I can do like a it's going to be the the final giveaway uh for the five day giveaway. I might do others in the future, but for this week we wanted to do like a since it's a big week in AI, we wanted to do uh like a giveaway. Sorry, Dave says uh we're testing coding or something else.

I'm just building some different applications just to test out GP5 and uh and everything. So that it might be some issues here in the actual um implementation plan. Might be some mock data here because it doesn't really build out the full app in one go. So maybe that's not perfect, but I kind of pushed it here just to make it complete. Hello Christian. Uh initial thoughts have seen some mixed reviews. Yeah, me too. Some mixed reviews.

Uh and to be honest, maybe a bit early for me to have any like big opinions, but uh it seems pretty good on tool calling and instruction following. So that is good for agentic work I guess and that is kind of my thing. So I think I could get some use of this Okay, so Claude code suggests we should do an AI voice narration dynamic background music generation for the app. So we want an audio pipeline.

It's a bit boring. Character consistency engine. Oh, that's going to be a tough one. Yeah, like uh someone said here, AI director mode, interactive generation. So it suggests here like a text stack uh AI for music composition ffmpeg. Okay. Yeah, pretty straightforward. So we are still working here in uh the IDE on cursor to build the app. writing out some main here. That might have been actually the issue on the running it on cursor agent because

I didn't really skim I just skimmed through the plan here. I didn't read it specifically. So there might be some issues in the plan here actually was so long 500 lines. Let's try it here with cursor now. So, I might as well for the giveaway today. It's going to be in like 30 minutes. Uh, we need to do a form because uh I don't have any quot yet on my uh YouTube API. Uh, I spent up

my daily quotota. So, again, we have to do a form like yesterday, Google form. So, I'm going to create one. Can you clear? Okay. Uh Can some people test the Can some people test the form here? Uh see if you can enter your name. Okay, so it's working. So people can enter for a giveaway. Okay great. So, if you want to join uh the giveaway in like uh 30 minutes, you can just start entering your name.

Uh it's just going to be the YouTube chat name. Hello, the Chucky. So, don't do don't do you don't need to do your real name. Just do your chat name or something. Okay. So, uh you can see now I think we are ready to test out the app here on in the cursor IDE now. So, we should be pretty good to go. We have our API keys. Uh we should just be able to open up this

in 3001. Okay, there's something Don't to do It's fine. So, let's do the same thing uh as we did before. Write a step by step guide to an intern how to use the app. Okay. Yeah, we have 31 responses. So, uh if you want to join, hello LS Arvin Bastian Hipley. If you want to join the giveaway, just type in your chat name. Okay. So, let's see the step-by-step instruction here for an intern. So, we're going

to start uh the backend terminal. Okay. We can do probably like this. Where does it start? Okay, so here it is. Okay, pretty good instructions here to be honest. Very easy to follow. We have the mpm run dev. Okay, so that was pretty good instructions. So, let's see the app now. Again, it's pretty boring. All right. Uh, let's do a high action street fight in Rome. But I just want to see if it works, right? It

did the image. Oh, what the hell is that? I don't know. Yeah, this is uh uh I don't think I'm going to do turn this into a video. I think I need to kind of understand [ __ ] a bit better first. We don't really get any comparisons. I think I need to experiment a bit more. It doesn't do anything. I think just the plan here we generated is just too bad. We'll figure this out. Yeah, I agree.

Edgar, what is that image? I don't know. What is it? Is it like a It's a URL here. Pick some seed. It's just picked a random seed here. Yeah, I think I need to I think I need some a bit more experiments with Gift5 before we Look at this image. It just found this. Is it a is a rock? Is it? Uh yeah, I think I'm going to skip the video. But I'm very happy with the

claude code app. That worked very good. I'm close to 60 million tokens in another 50 and I will master it. I hope to win says SQL. I nearly need cold code. That's pretty cool. Do some foreshadowing. Maybe you can uh um Okay. Is there any Don't people You can put in your name if you want to join them. Uh like the video too if you kind of helps me out a bit if you like the video.

Doesn't matter too much, but if you want to like it, that's fine. GPD hacked our outdoor camera. Okay, that's pretty cool. I did some uh I did some uh I did like uh I did this uh what do you call it? I can show you here. I did some interview and they kind of gave me this gift card. gift card and I I ordered one of these. Has anyone of you tried them? I ordered one of

these flipper zero. So, it was about $250. So, I ordered it today. So, I'm looking forward to playing around with it. I'm gonna try to write some custom code using uh GPT5 or something. Maybe write some custom code on this flipper zero. A customer of mine has uh shown it to me. It's scary. Yeah, I kind of want to buy one before it gets banned. So, I got this master card gift card I had to use

for something. So, I'm looking forward to it. It was a bit of a long delivery time though, up to two months. Yeah, it's a bit of a hacker kick. Uh, I really enjoy uh I don't know how many of you like this. I haven't really done this on uh I haven't really done this on stream, but I might think about doing a stream soon. I really enjoy this playing around with this page in my off time,

right? I don't know if some of you like this stuff. It's pretty fun. It's not sponsored or anything, but I kind of like to play around here. Pack a prompt. It's pretty fun. You can open cars, garage doors. Yeah, I know. So, I'm looking forward to try it out. Uh, they hacked the light up bracelet on entire audience or Taylor Swift concept. That's pretty crazy. John says he likes to get one, too. Sonnet over GP5 confirmed.

We'll see. Uh yeah, I'm looking forward to getting this. So, I thought maybe next week I might do like a hacker prompt stream. Uh it would be really fun to try to solve some of the the challenges together if people have some good suggestions. Could be pretty cool. Uh I really enjoy this page here. It's really fun. So, I'll definitely go check it out. So, we didn't kind of get like a a good conclusion today on

um any we didn't really get any good conclusion on the testing out the different uh models. Uh but it's really nice to hear what uh what people think so far. Some people like Edgar has uh has used it a lot more. So I guess I need to kind of dive in into [ __ ] Uh, I need to I need to kind of do a better job than this if I'm going to do a video on it because

this was horrible. It's not fun to watch a video when where nothing works. So, that's just the way it is. And I can't really blame I kind of did exactly the same on cloud code. Then same instructions, right? Except we use plan mode. Yeah, sorry Dave, you can join. If you see the link, it's uh yeah, it's just a cloud code probe subscription one month or JPT if you want to do that. That's up to you.

Or cursor. You can do cursor maybe. So, we have 41 responses. 42. I feel close code has turned completely stupid at certain times of the day. I made a spin a win simulator to pick a winner with RGBT and it works fine. Yeah, we did a spin a wheel yesterday. So, let's do that again. That was pretty fun. So, I'm going to load up chat GBT here. I'm going to uh we need a list first. Cloud

code is tired on Fridays. Hello. Uh Christian, is GT5 worth the hype? I think it's pretty good. Uh is it worth the hype? It's kind of hard to tell yet. Here you can add modules to do some really interesting stuff on the flipper. Yeah, I'm looking forward to it. When I get it, I might do like a video where I use Gift5 or Claw or something to do some C programming. I think you can do Python.

So, that's going to be pretty cool. What does it cost here? So, here is $1.99, but I have to I had to buy from like a Norwegian distributor, so it was like $250. So far, it hasn't really accomplished much. Uh, based on the content, it hasn't been worth the hype. No, I agree. But I need to at least do the weekend. I'm going to try other projects. Some projects I've done before maybe compare it. You can't

share links in the YouTube chat. I think it's just some um it's just uh it's like a security thing. The hardware I think it is. Is it a raspberry? Can't remember. Flipper zero. ST link developer board wifi. Not quite sure. Form factor looks too small. Yeah. Haven't really dived into kind of the hardware yet. Don't just abandon a tool because it only works for a certain job. No, I'm going to keep trying Gyp5 over the weekend

ESP32. Okay. Proprietary architecture. Cool. I'm looking forward to it. I'm going to show it off when I get it. Try it out. I saw some post on Hacker News today. I think it said uh where was it? Can't read this here. So, Flipper Zero dark web firmware bypasses rolling code security. So, I thought I might watch this video later. So they probably updated the firmware to kind of look at target vehicles on this rolling code uh

protocol or something crazy. Do you think open code can be a good uh placement for claw code? I think Edgar has a lot of experience with open code. I have just started using it. Uh but so far I kind of like it. What I like about it is uh it's very easy to use models. You can just pick they have GD5, they have the open source Kim K2, Horizon Beta, uh Quen 3. So they have a

lot of models. So it's very easy to test out new models using open code. So, I kind of like it so far. They have I was thinking maybe I should try a GT5 Mini. Uh, let's give it a go here. So, uh, but not here. Let's do Let's do GP5 mini um what I call it GPT GTP 5 minute open code and let's do uh models GPT file and mini and let's do uh create uh create

a HTML of two black holes merging. Pick the best tech stack. Save it. Save it as BH2.html. Okay, so this is kind of open code. So, I want to try the GT5 Mini. Uh, I like all the models with open code. That's pretty cool. Hello in raw. Which one is better? Uh, I can't I can't say I really have gotten much out of GD5 yet. I need to test it out more. I think it's going to

be pretty good, though. So, what I'm looking forward to with GD5 is I'm going to try to do some MCP stuff this weekend. Uh, because it's really good at tool calling. I think it could be really fun to try it out with MCPS. So, I'm kind of going to look into that this weekend. I think that could be pretty interesting. Okay, so we ran uh GPT5 mini on the two black holes merging. We're going to do

a WebGL shader. So, let's see here. Okay, at least it did work. Wasn't that interesting? It was quick though. Okay, let's do uh good do a pelican riding a bicycle SVG. Save as P.SVG. Most of your tests uh botfather fail and claude just uh gets me much more better and writes cleaner code. No, I agree. JK software that wasn't too good. We have much we have made much better than this. We did some of these tests

with the horizon. That was very good. Yeah, I think over the weekend we kind of can look at the sentiment and I guess more people had time over the weekend to try out GD5 and like GT5 Mini and the other models uh just to see. So, let's check out the Pelican SVG because it has some errors. through more tests needed so far. Has any of you guys have some more scenarios where GD5 is uh more appropriate?

I haven't, but I know other people here have tried it out on more uh real use cases like uh code bases and I haven't so I can't really comment on that. Okay, here is the bicycle SVG from Gypin. It's animated. I don't know why I didn't ask it to do that. Kind of missed the bicycle though. The animation is fun. How do I do this? Okay. So, it's going to be 2 minutes until the drawing. We

have 54 uh responses. So, I'm going to put up like a timer. So, if you want to join the giveaway for a pro chat subscription cloud code, it's up to you. So, I'm going to set the timer 2 minutes until we stop 2 minutes. Any reason that everyone is just has HTML? Uh, I don't know. It's just very easy. If I'm going to do like a big test like in Cloud Code, I've tried some uh testdriven

development test, but it just takes hours, doesn't fit so well into a stream. What are the chances that GP5 thinking is just 04? We will see. Said I don't know. I don't think they're going to name their models like 04 and 03. I think that time is done now. None of these of the models are good with the Rust language. I guess it's just not enough training data. That's probably it. I'm not quite sure. Uh, okay.

One minute for the giveaway. Ash, hello. So, you can join there. We have prepared. We're going to prepare a prompt here. So, I'm just going to do extract the full list of names from the CSV because I had a lot of issues yesterday trying to do it in uh one operation. Okay, this was a bit better. This is the SVG Pelican from uh from uh GPT5 Mini. Don't give it. Hello S is on. Okay, last chance.

So the two minutes has passed. So let's grab the form 64 entrance. Yeah, it was not too bad for GT4 Mini. I heard people said it's pretty good. Okay, the two minutes has passed, so I'm going to download this. Oh, someone got in really late there. Okay. Uh, I think I need to unzip it. Okay. So hello Paulo can't do this. I just updated GP on Android. Still no GT5. Okay. Master Snake. I have GT5 on

my iPhone at least. But my friend, he doesn't have GT5 on the desktop, only on his computer. Okay. So, we have the list Uh remove dupes and print list in uh print list separated by uh okay so we have the list here. Okay. So, we have a list of names. I'm going to give it to you. So, what you can do now is create a spinning wheel in React. So, the wheel is going to have kind

of the name of all the usernames uh from the list. And when we we're going to spin the wheel to find a winner, uh, of course, uh, it's going to be random. Uh, but the wheel has to kind of, uh, show who the winner is. That was a bad prompt. Here is the list. Um create the simple app. Add a sound effect. There's going to be a lot of lines here. Pick a uniform random window index

and compute the target rotation to land on it. Okay, so it's going to be pre-drawn, right? But we're not going to see that. So that's good. Okay. So, let's do this now. So, good luck everyone. I think we are about 70 people something like this. So, good luck. Drum rolls. So, we're only going to spin it once. And if um yeah, the UI is not so great. Uh but I don't care. So, let's spin it once.

And let's see. Uh one, two, three. Okay. Uh, yeah. Again, it kind of failed on the but the winner is Barnaba Barcelona. If you're here, please let us know. If not, we're going to spin again. Barnaba, are you here? Yes, Barnaba is here. Congratulations. So, what you can do, send me an email and we'll figure it out from there. You confirm. Can you confirm, Barnaba, you got the email? Yeah, no problem. We'll figure out on email

how I can chip it. We're going to do other giveaways. Uh, but I need to come up something interesting. Uh, sorry Dave. I think I did this all week and I've only got one guy trying to claim their price, but uh, I figured that out. There's no problem. I'm used to getting like I get maybe like 20 scam attempts on my YouTube channel every day. So I'm pretty pretty um used to that. Okay. So in raw

want to test this luck. Spin it one more time. We did that yesterday too. Okay. So it wasn't you. It was someone else size but that didn't count. Okay. So, um Barnaba send me an email. Radio says, "How good GPD5 guys?" Some people have good experiences, some people are a bit meh, and I'm I think I'm a bit too early. Yeah sorry Sai. Maybe next week. I'll do some giveaways next week, too. So, I think there's

kind of mixed opinion so far on GD5, but uh I think it's going to be good. It's competition is always good for the consumers. So, if GPD5 can push on CL code, that's just good, right? So, I welcome the competition, but I don't think I'm going to give up cloud code yet for the C square C code. Uh C square uh C code I write is quite good. Okay, that's cool. Yeah, it's faster, but it's kind

of hard to tell if you're in the application. It's kind of hard to tell if uh if you are kind of in the chat app which model is it picking. If I say hi, it's very fast. So that's cool. But I can also it kind of picks the model. Theas have some good experiences with it on cursor max and run the best model was trying not to oneshot anything but progressively make changes. Uh we have a

strong community. The chat window will testament that. Yeah, that's pretty good. And it's going to be interesting next week to hear how people have used like GT5 and the sentiment. I think that's could be pretty cool to hear. I'm going to do some testing myself this weekend and maybe we have um another opinion next week. PHP plug-in creation task GP5 has been good. Cool. I'm going to watch some I did I saw AI explained uh Phillip

he had a video on GT5. I thought it was pretty good if you haven't checked it out. I kind of agree with him. He kind of went the angle that GPT5 is very good for most people since a lot of people was stuck on GPT40. They didn't bother changing the models up here, right? So like my sister, she she used chuck GPT, but she never picked models. She was always on 40. And um now she's she

can be on like GD5 instead for some queries I guess. So overall I think this is a good people for most good change for most people but for us like early adapters API users we kind of maybe expected a bit more but we'll see. GP3 is good for most people too I guess but now everyone is kind of they don't have to worry about the picker for what will you use flipper uh we'll see Sanjay I

have some plans but yeah uh I think I need to go uh I always pick my model specifically for conversation Yeah. Uh, I agree, Michael. I I wish I could had 03 model here and stuff, but like my sister and the most of the 700 million users of SHA GBT, they probably don't pick models. They just go with the standard here. Yeah, Richard, thank you for tuning in. Ken, thank you for tuning in and I wish

you all a great weekend. try out some models and let's hear next week what you think. So, thank you everyone for tuning in. We didn't get so much out of the stream, but uh we had some good conversations. Edgar, as always, thank you for tuning in. Ken, nice to see you back here. I'm looking forward to hearing your opinions next week. Yeah, I agree. Christian Opus 4.1 more useful for coding. Yeah, it's going to be interesting

to see. Have a great weekend. Thanks, Sai. And uh David, have a nice weekend. Paulo, bye-bye.