I've been testing GPT-5.2 for a while now. It's a serious leap forward in complex reasoning, math, coding, and simulations. It built a full 3D graphics engine in a single file. Interactive controls. 4K export. One shot. The pace of progress is unreal.

It’s also the best agentic model OpenAI has shipped, runs tons of tools in a row without issues and is faster than its predecessor. I built an agent that responds with GPT-5.2, 5.1, and 5 at the same time, 5.2 calls tools with no preamble, and doesn’t get lost in long sessions.

In the video above, I used GPT-5.2 thinking on high reasoning effort.

Also look at these benchmarks!

Nice cook! Both u and gpt. It seems this week is a big week cuz we also launch an open-source model which means everyone can use ai to control their own phones. Here's the link:

so the exponential was real. damn.

As it was told.

Are you certain it didn’t use some library? E.g. its nuts what you can one shot with three.js, but that’s still far from building an engine from scratch.

All code and shapes from scratch.

In a scale of 1-10 how fried is your gpu

11

This is just in training data

Bruh... Why you be holding out. 😂😂😂

would u say it gets close to opus ?

Better than opus in a few ways! I can’t even believe I am saying this.

I meann, if you're saying this i must try it out asap. Ty 🫡

one shot full 3D engine with controls and 4K export hits different. thats not just iteration, thats a capability jump in reasoning depth

TBH that looks terrible

can you give me prompt?

We get so much of this bs on x. When you try it yourself it never works. I don't belive you.

Wtf why is it able to do this for you but I ask it one simple task and I have to fight it for 2 hours on edits I didn't ask for which end up breaking logic?

Bruh, is it building my next vacation too? 😂

what was the prompt you used for this? it’s obviously a really good model, but i’m not seeing quite that insane results myself for my own work (i still have to steer it; it does think strongly for itself as a senior colleague would)

Do you mind dropping the prompt bossman.

"One shot" is the flex but let's talk about reproducibility 😏 Yes, GPT-5.2's extended context window (400k tokens per OpenAI's release notes) enables complex single-file builds like this 3D engine. But anyone who's shipped production code knows the real test: can it do this consistently, or is this cherry-picked demo magic? According to Fortune, GPT-5.2 shows 15-20% improvement in coding benchmarks vs 5.1. Impressive, but we've been burned before by "unreal progress" that falls apart when you ask it to refactor legacy PHP. Still, this is genuinely wild work 🔥👏

Your browser is crying

Pero dile que realice el codigo en python para un simple semaforo simulado y mira el resultado pesimo.

AI still can't build shit past a simple prototype. Ask it to edit a project with 10k+ lines of code it will shit itself and write the most atrocious slop you've ever seen

This pace of progress is making my head spin

either you're in the most exclusive beta program on earth or you just pulled off the best tech demo flex of 2025 either way that 3D engine is nuts

No. Ask it to make anything that required emotion and it flops. •

May I know the prompt please as i would like to testother model pls thx

OMG! Not so fast! 😅 Incredible results!

The pace of progress is unreal and it seems like it's sped up the last two months..... CRAZY

Wow ✨

Looks cool

Does it beat gemini3pro for front end?

@grok what promt he might have used

Get back to me when you make a 4D graphics engine

The final output seems impressive

And then nothing,internet is full of those projects

Good to hear thar