Same prompt, 9 months later

Wow incredible progress! GPT obviously looks the best but the boxes pop out of nowhere. Fable is the least buggy. Grok is has the most bugs still. I kind of like the Gemini one. The moving arm has quite the nice bounce to its movement

yeah! they all approached it so interestingly

Fable looks best, Astra got the perspective wrong on the boxes and overdid it imo

Bro, I just made a Gemini 3.8 Flash in Agent Hermes.

that looks amazing, i wonder how GPT-6 tackles that

If you're interested, bro, we can test one promt on different models and systems. Tell me which promt I'll try, test it, and show you the results.

'Create an interactive 'Tearable Cloth' simulation using HTML/CSS/JS in a single file. Render a grid of points connected by constraints (springs) that react to gravity. Allow the user to drag the cloth with their mouse and right-click (or hold Shift) to slice/cut the connections.'

Gemini 3.8 Flash Hight worked for 7 minutes in /goal mode.

pretty cool

yes, it seems to be good, but I need to work on agent md and sould md, and then it will definitely be better, I created a new project in Hermes agent, so I feel that the potential was not fully revealed

Grok 4.6 looks extremely good for its size, fable and astra are 10x its size almost

lets see grok 4.7

These type of benchmarks I care about

🫡

Everyone looking st astra. But gemini did pretty good for a flash model, and doesnt blow up a 5h limit in like 10 mins.

gemini was incredibly fast, it was crazy

Im pretty sure that the main thing wrong with Gemini when people try to use it is that it is stuck inside of antigravity (usually) which is why gemini high can compare to sol high on benchmarks, its not using antigravity in the benchmarks.

Nine months between the same robot-arm prompt is such a fun time capsule. Have you got any other old prompts you’re tempted to rerun?

yessir, a lot of them I will try to remake all the experiments from my paper

The rocket launch one too 👀 Looking forward to seeing those experiments rerun. Nine months already gave us quite a jump.

That much progress in 9 months?? That's insane

imagine 10 years from now

😂 only type of benchmark that shows truth. IMHO Gemini is the best in arm movement, logic, & blocks’ persistence. Astra looks impressive at first, but the sentence fragments trigger me

true

I keep one prompt of my own for this, a dull formatting job I rerun every few months. Same input every time, which is the only way I have noticed a model getting worse.

wow, interesting results

Nothing dramatic. Same input, and it started explaining the task back to me instead of just doing it. Yours went the other way, which is the nicer surprise.

Fucking made a track from the whole thing too

🎶

nobody asked for a pause button

The jump in quality in nine months is wild

Gemini had the best result here. Let go of of the hate. Gemini > Claude > Grok > ChatGPT

Fable 5.1 high not ultra why

Fable 5.1 free via is is high effort

Gemini 3.8 flash looks the coolest

it was for sure the fastest to generate, lighting speed

The jump in Chat Gpt and Gemini is unreal.

3D(i) has been achieved!

and when it uses blender natively is even better

publish the song bro

music also a certified banger

🕺 🪩

Wow astra is amazing.

right imagine!

Gemini najbardziej logiczne

Qwen3.8 Flash Next (UD-IQ4_XS) run locally. same prompt one shot

Nice one!

Progress

love it

9 months after prompting.. O.o

Astra is truly the holy grail of 3D

True

Same prompt 9 months apart is the fairest benchmark there is. The difference now is the model executes a shot list instead of guessing one.

we have come far ahead

Michelangelo meets AI in a stunning new form.

Tried something similar

same prompt 9 months later is a great test

Thanks!

🤝

The four results are impressive, the progression is obvious. Astra delivers a very "design-focused" result (a little too showy), while Gemini 3.8 Flash High and Fable 5.1 High are interesting. But Grok 4.6 xHigh impresses me the most, because it follows the prompt "Make a looping HTML animation..." most closely. It reuses the pieces that were moved and puts them back in their original positions, whereas the other models make the pieces appear (or disappear) before moving them. My ranking is as follows: Grok 4.6, Gemini 3.8 Flash High, Astra Ultra, Fable 5.1 High. I’m surprised by this ranking myself. I was expecting more from Astra, but one prompt doesn’t make a model...

visually gpt is the best, but fable is physically more accurate and the visuals arent bad either.

The approach from 4.1 thinking was hilarious.

Once again Gemini mops up

the jump is nice, but how much is the model and how much is the prompt?