Transcription
[Music] [Applause]
Hi, welcome to another video. So, OpenAI has launched their new GPT-5 CodeX model, which is an improved version of GPT-5, specifically enhanced for agentic tasks and coding, especially in CodeX. This model is now integrated throughout CodeX tool, Codec web, and other tools like that. They basically say that this model is better at coding, costs less since it uses around 90% fewer tokens while delivering better performance, and is also faster for most tasks, which is kind of cool. It can now also think for longer on complex tasks as well.
So, as you know, I've been working on some agentic tests. It's still just at four tasks. I'm working to increase that, but these four are good benchmarks to test the stuff. Currently, CodeBuff is awesome and is really good, but it costs a lot. Almost more than double the price, which isn't a good thing, and I can't really use it much. But let's see if GPT-5 CodeX is actually useful. The previous GPT-5 CodeX was the lowest scorer for me. But now, let me show you how you can use CodeX and the results it gave me.
But before proceeding, let me tell you about Ninja Chat. Ninja is an all-in-one AI platform where for just $11 per month, you get access to top AI models like GPT-4o, Claude 4 Sonnet, and Gemini 2.5 Pro, all in one place. I've been using Gemini for quick research. But what's really cool is their AI playground where you can compare responses from different models side by side. Their mind map generator is a game-changer for organizing complex ideas as well. The basic plan gives you 1,000 messages, 30 images, and five videos monthly with higher tiers available. If you need more, use my code king25 for 25% off any plan or king40 yearly for 40% off annual subscriptions. Check the link in description to try it yourself.
Now, back to the video. So, using it is simple. You just get CodeX upgraded with a pretty straightforward command, and you should be good to go. If you have a ChatGPT subscription, then you can use it with that. They've lifted the limits now, and it ranges from about 30 to 150 messages per 5 hours on the $20 plan. I don't really get such a huge margin here. I mean, 150 is like 5 * 30, which doesn't make sense to me. Probably you'll mostly see 30 messages because I think it will be tuned based on load, and right now many people will be trying CodeX. So yeah, I don't really like these uncertain limits. It's still worse pricing than GLM Code, but we'll see. If you're a pro user, then you'll get about 300 to 1,500 messages, which is again a really wide range, and I don't know what exactly this range is based on.
Anyway, let's go ahead and start using it. So, it works the same as before, and you can check out my Deepseek CodeX video where I go in-depth about how it works and how you can hook it up with DeepSeek. But that ain't the video for today. If you open it after a while, it shows you this animation for the new GPT-5 CodeX, which is interesting. Then just hit try, and you should be good to go.
Now let's try our first question, which is to ask it to make a movie tracker Expo app using the TMDB API. The previous CodeX was pretty bad at this, so let's see how it compares. This is what it made, and it's kind of good. It has a different style altogether, and I'm not sure if I like that or not because it doesn't look very much like how movie tracker apps usually are. It has a bit of a different vibe, and I'd say that's good, but not as good as what CodeBuff or Claude makes. But design is subjective, so you may like it. It also doesn't have a movie details page or features like that, which is not great. Claude makes something better, which is generally more in line with what I want. CodeBuff is the best here and really solid for sure. Previous CodeX with plain GPT-5 was amazingly bad, but this one is amazingly good.
Moving to the next question. This was another weak spot for CodeX, which is to make a visual calculator in the terminal using Go and Bubble Tea. And you can see this is what the new GPT-5 CodeX makes. It's actually quite awesome. Honestly, it's one of the best I've seen yet while using extremely few tokens. So yeah, this is awesome.
Now after this, we've got something that I think was knowledge-limited in at least the last GPT-5. That's asking it to edit an FPS game made in Godot. I think only Opus and Sonnet really know how to write good GDScript. So, is the new model able to make good games? Well, the answer is no. It still can't really work with Godot and throws a ton of syntax errors. It tries to write a Python script, which is different from GDScript, and that just breaks it. So, yeah, still bad at Godot.
Moving on, I tested it on editing the open code repo to add an SVG creation modal, and it failed here because multiple files need to be edited. None of them passed apart from CodeBuff, but CodeBuff takes a lot of tokens. So yeah, that remains a problem, but I'd say it's now better than Claude Code and takes the second spot right under CodeBuff. And that's what I feel too. It slightly beats Claude Code and for a bit less money if we factor in the GPT-5 API pricing, assuming this costs the same. So that's kind of cool. I find it really awesome. I just hope they keep the limit around 300 messages. That would be awesome.
I still think that the $3 GLM coding plan is unbeatable for students or people on a budget. I have a video coming up about it where I'll show you how to maximize that plan. But this is also great, and I hope the API for this gets released too so we can use it in tools like Rode or Klein, because I honestly prefer those over CodeX. I like the CodeX VS Code extension since it's less memory-hungry than RU, but it still doesn't match the raw performance you get out of RU. Combining CodeX with RU would make a strong alternative to Sonnet for sure. Let's see what happens. And maybe we'll even get a GPT-5 mini CodeX variant that costs less. I really like the smaller GPT-5 mini model. So, let's see.
This one uses way fewer tokens, costs less, and performs better, which is exactly what everyone wants these days. I'm glad OpenAI is actually listening and building useful stuff instead of just benchmark maxing. Overall, it's pretty cool.
Anyway, share your thoughts below and subscribe to the channel. You can also donate via Super Thanks option or join the channel as well and get some perks. I'll see you in the next video. Bye.
[Music]