Pinned Tweet
GPT-6.2-Astra vs Gemini 4 Neon
Pure Three.js. 20-second dawn walk through a rain-washed Kyoto Senbon Torii path and Arashiyama bamboo grove.
GPT-6.2-Astra (Top): $6.67 · 41m 9s
Gemini 4 Neon (Bottom): $9.14 · 1h 8m 19s
Astra or Neon? 👇♥️
GPT-6.2-Astra vs Gemini 4 Neon
Pure Three.js. Same prompt, 20-second sunset gondola ride under a Venetian stone bridge onto the Grand Canal.
GPT-6.2-Astra (Top): $11.78 · 1h 9m 24s
Gemini 4 Neon (Bottom): $16.63 · 2h 0m 16s
Astra or Neon?👇♥️
7
19
972
Unreal Engine 5? No, pure Three.js written from scratch by GPT-6.2-Astra vs Gemini 4 Neon.
Same prompt, 20-second walk down a Gothic limestone cathedral toward its stained-glass rose window.
GPT-6.2-Astra (Top): $7.15 · 40m 31s
Gemini 4 Neon (Bottom): $8.41 · 1h 3m 40s
Astra or Neon? 👇♥️
GPT-6.2-Astra vs Gemini 4 Neon
Pure Three.js. Same prompt, 20-second Monaco GP tunnel exit into Port Hercule at twilight.
GPT-6.2-Astra (Top): $6.77 · 41m 35s
Gemini 4 Neon (Bottom): $7.26 · 55m 44s
Astra or Neon? 👇♥️
This is NOT a game engine. This is NOT GTA 6.
One day after GTA 6 launched, Google opened up the Genie 4 API, and I just drove a 20-second neon Miami chase generated live at 24fps with native stereo audio.
At 0:05 I injected a live text prompt for police lights, and a flashing cruiser immediately pulled up beside me.
We are literally steering a hallucinated reality in real time. Wild.💀
Total cost: $2.05 (~$0.10/sec).
GPT-6.2-Astra vs GPT-6.2-Sol vs Gemini 4 Neon
We pushed 3 frontier models to nail the hardest part of the Burj Khalifa in Three.js: the upper spiral setbacks, louvered mechanical crown, and a full 360-degree summit orbit.
GPT-6.2-Astra (Top): 1h 11m 16s
GPT-6.2-Sol (Middle): 1h 16m 44s
Gemini 4 Neon (Bottom): 1h 25m 18s
Which one wins?👇
Testing Google's new Gemini 4 Neon on... actual neon.
GPT-6.2-Astra vs Gemini 4 Neon
Same prompt: one continuous 20-second Three.js steadicam shot through a rainy Tokyo alley, across a wet intersection, and up to an elevated train.
GPT-6.2-Astra (Top): $6.48 · 46m 56s
Gemini 4 Neon (Bottom): $6.40 · 48m 5s
Same $6.40 budget.
Which Tokyo night are you shipping?👇
One prompt. Zero edits. Around 10 minutes each.
I asked GPT-6.2-Astra and GPT-6.2-Sol to one-shot a Three.js North Sea oil rig getting slammed by a rogue wave.
GPT-6.2-Astra (Top): $0.99 · 9m 39s
GPT-6.2-Sol (Bottom): $0.28 · 11m 13s
28 cents for that storm lighting and flare stack is wild.
Which one takes it: Astra at $0.99, or Sol at $0.28?👇
5
15
1,333
GPT-6.2-Astra vs GPT-6.2-Sol
Slow-motion Three.js cheetah sprint test!
Quadruped leg biomechanics expose every flaw.💀
Same prompt, same xhigh effort:
GPT-6.2-Astra (Top): $13.85 · 1h 54m 8s
GPT-6.2-Sol (Bottom): $4.03 · 2h 24m 58s
Astra finished 31 minutes faster and nailed the stride!
Worth 3.4x the cost? Which one would you pay for?👇👇👇
GPT-6.2-Astra vs GPT-6.2-Sol
OpenAI just dropped both, so I ran them head-to-head on your favorite Three.js octopus test.
Same prompt, same xhigh effort:
GPT-6.2-Astra (Top): $10.30 · 2h 22m 15s
GPT-6.2-Sol (Bottom): $3.65 · 1h 28m 24s
Which one would you pay for: Astra's $10.30 detail, or Sol's $3.65 run at nearly a third of the cost?👇❤️
Claude Fable 5.2 vs Gemini 4 Carbon vs GPT-6-Astra vs Gemini 4 Hydrogen
Same prompt. Build a 20-second Three.js humpback whale breach off Icelandic basalt stacks, from bursting out of the swell and rolling mid-air to a broadside slam and final tail lift.
Claude Fable 5.2 (Top left): $19.07 · 71m29s · xhigh
Gemini 4 Carbon (Top right): $13.11 · 163m18s · high
GPT-6-Astra (Bottom left): $11.10 · 72m45s · xhigh
Gemini 4 Hydrogen (Bottom right): $4.27 · 37m27s · xhigh
Fable 5.2 runs 4.5x Hydrogen's API spend for better lighting and a cleaner final beat.
Which one would you actually pay for?👇
Feels like a storybook town I'd happily get lost in.
Didn't expect the visuals and tiny details to be this sharp, and the soundtrack honestly seals it.
ngl the three.js shot lowkey clears the blender one
how is pure code looking better than a full 3d render lmao
7
13
1,022
Three.js + coding agents is such a good pairing right now.
The pure-code setup gives the model room to actually cook, and the agents keep raising the bar for what Three.js can look like out of the box.
2
4
333
Hydrogen barely just dropped and Google's already stealth-testing Oxygen on Arena.
And from this clip, it clears Hydrogen by a pretty wide margin.
Google is cooking!
wild how one of these chameleons took almost 4x longer to generate than the others.
same Three.js prompt, three models, all on high effort: a chameleon striking a beetle off a mossy branch.
Gemini 4 Carbon · Gemini 4 Argon · Gemini 4 Hydrogen
Carbon went all-in on bumpy, color-shifting scales, but the tongue completely sags. Argon nailed the gripping feet, then turned the tongue tip into a weird red blob. Hydrogen kept the low-poly jungle and strike clean, and finished 3.9x faster than Carbon.
here's the catch: slower and heavier doesn't automatically mean a better animation.
which one actually looks worth the compute to you? Carbon, Argon, or Hydrogen? drop your pick👇
Making Siri the flagship selling point of a brand new hardware category is a bold strategy.
Smart glasses are obviously the right form factor for AI, but Apple really needs Siri to actually work reliably😂
NEW Power On: Apple’s next big product — smart glasses planned next year — will offer Siri AI app integrations as its flagship feature. The idea: Book an Uber, reserve a table or order a coffee from your Apple Glasses — with apps running on paired iPhone. bloomberg.com/news/newslette…
3
221



