Gemini 4 Carbon just redeemed Google on 3D for me. Did same test with Argon few weeks ago and left highly disappointed. But with Carbon it really delivers. Ran both tests in Antigravity on High effort. Gemini 4 Carbon: 81 min, 186k output tokens, about $4.5 in API terms.... Gemini 4 Argon: 14 min, 11k output tokens, about $1.8 in API terms. Made with Higgsfield plugin in Blender.show more

Big difference!

That's a massive difference. Carbon is going to put people out of work

81 min vs 14 min and you can see where the time went: the Carbon dragon has way more detail in the spikes and wings. Love that you posted the token count and cost next to it, that's the part people actually need to decide.

Holy, Carbon is different.

I'd run that same prompt five more times before calling Google redeemed. single runs on 3D are mostly luck

funny how carbon is the cheap one per token but your run cost 2.5x more because it wrote like 17x the tokens. that's why i only look at cost per finished task now, the price page and the invoice tell very different stories

I’ve just been going seedance 2.5 and kling for making realistic 3D video sequences of nano 2.1 renderings.

它太慢了

Same test, same harness: Argon flopped, Carbon delivered. That's the before/after I trust more than any leaderboard. One person, one task, two runs.

17x the output tokens for 2.5x the price, I'd take that trade on 3d stuff every time

186k tokens for a 3D model is one expensive cube

17x the output tokens for 2.5x the price, and a flop turned into a keeper. Cheap trade I'd love to see where the extra 67 minutes went

running the exact same test on both is what makes this worth reading. most “X is back” posts change the prompt too. where did Argon fall apart last time, the geometry itself or the materials and lighting?

Finally a Carbon 3D run that doesn't end in disappointment.

Same test, same harness, different verdict, glad you kept it apples to apples

Agreed! Just did my first run on it.

Running the same prompt on both models makes the comparison honest. The leafy seahorse fins sold it for me.

卓越的提升～

Woahh look at the detail 😍

81 min and about $4.5 for a test that used to let you down is wild. is Carbon good enough now that you'd switch your 3D work over, or still testing?

hi, stefan, pls check your gmail inbox, so thx

What is the prompt and how did you do the animations with blender? Were the animations also generated with the plugging inside blender

$4.5 for 81 minutes of agent work is cheap. i'd want to know how many of those 186k tokens went on retries vs actual building

same test that flopped last time and now it lands. that's the only benchmark worth trusting tbh

八十一分钟烧掉十八万 token，这速度简直像开挂！😲

81 minutes and $4.50 to feel redeemed. Argon did it in 14.

How does it compare to Opus in your opinion!? Worth the switch?

Same test, same tool, same effort level is how these comparisons should be done. How did the 3D output look up close?