Prompt: Create a rocket launch animation starting with engine ignition and heavy smoke, followed by a slow liftoff that accelerates, using HTML/CSS/JS in a single HTML file.

I tried with the GLM 4.7 version. 534 lines of codes. It took 4 minutes. I'm impressed.

@thanhtran32 Oh wow that looks good!

GPT 5.2 best.

GPT-5.2 won this one!! It was very very wood, even the camera shakes

opus acts like it is the only option and nothing else matters

Right!! Opus is very interesting even tho in this test the specific part of the lift off can’t be seen, the level of detail it puts is wild to me

Rocket liftoff simulation HTML HTML Output Lines: 618 lines of HTML (gpt-5.2-extended-thinking) 332 lines of HTML (gemini-3-pro) 188 lines of HTML (grok-4.1-thinking) 820 lines of HTML (opus-4.5-thinking-32k) Model Details: gpt-5.2-extended-thinking: Score 9.6/10. Reasoning Time: 22 s, Total Response Time: 129 s, Total Reasoning Words: 205 gemini-3-pro: Score 9/10. Reasoning Time: 10 s, Total Response Time: 36 s, Total Reasoning Words: 203 grok-4.1-thinking: Score 2.8/10. Reasoning Time: 6 s, Total Response Time: 10 s, Total Reasoning Words: 63 claude-opus-4.5-thinking-32k: Score 7.4/10. Reasoning Time: 72 s, Total Response Time: 184 s, Total Reasoning Words: 1353 *All experiments use the model’s very first, unrefined output, without retries or additional guidance. *Github with the original code and more stats: * The reasoning time was measured manually starting when the return key was pressed. In the images, the timer starts only when the model begins inferencing, and there is a short delay between these two events. This is why the times may not exactly match.

gemini cook on this one im telling you look what it added.

Oh wow! 1 shot?

this one added sound and everything and birds

Love it! Opus would have been amazing but messed it up at end haha. Still good efforts in creativity as always from Claude. Gpt-5.2-ex quite amazed me too. Here is Gemini 3 PRO + Deepthink (417 lines, quite smooth yet simple):

Nice very smooth!!!

nice graphics! 1 shot?

Almost 😅 The first shot had a bug; this is the result after the second comment. You can check the previous results here: (The first link should be to the chat; I messed up.)

Pretty cool

this is #glm-4.7 with cc and ultrathink 🤠

That’s a good one

I dont really like these tests because each model has their own prompt requirements. You have to sit with a model to know what prompts you will need to force outputs, so jamming some universal thing into them isnt that impressive to me. I dont think we even know how to benchmark models, I doubt even the singularity would know. We are just caveman dancing around the fire.

Hey this is alien technology, we discovered how to mimic human intelligence multiplying mathematical matrices. But the purpose of this test is to try to mimic how a user will vibe code on the web UI and analyze performance, reasoning, inference time and social media impressions

Yeah thats fair, I just thought you made gemini 3 look like dogshit so i was flaming you a little bit lol, my bad.

all good lol, here is a good one gemini did great

AGI is here

@snapss Opus is gooood

Gemini 3 Pro

Nice

Dude right! They have All the space x training info

WTF happened to Grok this time 🤣 Opus on its own lane

@businessgrowth Grok did soo bad haha

Minimax M2.1 here, one shot, 628 lines

Very cool

Opus, what happened are you okay?

This is so cool , had to try this.

Very nice!

this toke a couple prompts but it added sounds and birds and some other things. 5.2 sucks so bad . i feeel like gemini is the best all round model. claude is good at coding but not by much on gemini

Thats pretty cool, on my test there are all first shotted but 1-3 more prompts can make magic!!!

nah opus got me wheezing

This is the new benchmark: not who answers… but who understands the vibe.

Love it. Especially how “the smartest model in the world” comes last on all these tests lol

@robb_schmidt 😆

I prefer GLM4.7 version 😊

Interesting

$ICP one shot

Lol opus is dumb

opus made a nuke lmao

lol right!! exploded

FYI 3D simulation w/ physics for balancing a rocket: Used Opus for the trickiest bits. Physics thx to @lorentzframe.

@lorentzframe Nice

I love these! How did you make the music?

the good old @suno

Same prompt with GLM-4.7

GLM-4.7 is underrated

@__elperro__ I think it's actually overrated rn

@diegocabezas01 Why do you think that?

This is the benchmark that actually matters. Not “who writes nicer text”, but who can build working dynamic systems — state, timing, motion, logic. We’re entering the phase where LLMs compete as software engineers, not chatbots.

2026 will be even crazier

I generated it with GLM 4.7 and it played a loud sound that almost made me deaf. That's dangerous.

Share!!

sound started good and then went wild lol

@grok wtf 😭 bro try again, gimme a better one.

@grok please

bro @elonmusk wtf grok?

@elonmusk 😱

How come no one is talking about the song?? Too good

Thanks!! 🕺

Interesting 🤔

🙏

Ironic about Grok and rockets :/

With all the space x info