Gemini 4 Pro vs Opus 5 (max) i gave both models same prompt to build 3D dragon using three.js the hype around Gemini 4 Pro is insane so here's a direct comparison against Opus 5 on max effort tbf Gemini does keeps up with Opus 5's performance which is... on max effort and that's impressive in itself, who do you think better? prompt used: create a 3D dragon using three.js, use white theme background so main object is visibleshow more

how long did both of them take in comparison?

didn’t pay much attention to it as both took similar time around 10-15 minutes

Where did you find Gemini 4?

it’s stealth released inside Arena battles

@grok do these types of videos show how sophisticated an ai is?

Y gémini 4 pro dónde lo probaste en tus sueños jajajajaja no existe aún.

It's good output from Gemini, the problem with Google model it's cannot remains in the competition for at least a week.

Both dragons came out fire Gemini 4 Pro keeping up with Opus 5 on max effort is seriously impressive!

in that specific task opus edges it out, but the progress gemini 4 made is genuinely insane think it's more useful comparing it to gemini 3.8 flash to actually see the real jump I did exactly that and ngl 3.8 feels like it's from 2023 while gemini 4 feels straight outta 2026

Esse design parece do claude.

apparently， the opus 5 is better

Both are so bad

While the dragon drawn by Opus 5 is an Eastern dragon, like those found in China, the one drawn by Gemini 4 Pro has a Western style. It might be difficult unless we use the same subject.

Opus is more stylized and looks better to me. Movements are much more interesting.

The fact that Gemini can stay this close to Opus on a 3D build is already impressive. These coding models are getting ridiculously competitive.

My Google ai pro plan is so backkkk! $GOOGL

The dragon comparison is a useful matched-prompt test. Add raw Three.js, runtime labels, interaction checks, and a second requested edit; then report repair rate and time-to-playable result so the Gemini versus Opus gap survives beyond the first render.

Gemini 4 is good!

Gemini's dragon actually looks better ngl

Model comparisons on 3D generation are becoming as common as coding benchmarks. That's the real signal. When frontier labs are getting graded on 'build me a dragon,' spatial/3D reasoning is clearly the next frontier. 🐉

The Gemini version looks like a plesiosaur.

Opus 5 is a terrible model though….you should also compare it against 5.6 Sol and 6 Astra tbf

CES DEUX LÀ SONT DES CRIMINELS. ASSASSINS

Nice comparison. I would add a second prompt with messy requirements and a small code fix, since polished demos can hide how models handle iteration.

The second prompt is the important test. A small fix under messy requirements shows whether the model can preserve the existing structure instead of rebuilding everything.