vibing through ai and backend sharing what I like ai tools and demos ✉️ [email protected]
Joined October 2022
- Tweets4.8K
- Following974
- Followers1.3K
- Likes2.1K
seeing a bunch of new people here lately so hey, i'm sou :>
i write code, build random stuff and spend way too much time on the internet
mostly into backend, ai agents and lately trying to figure out indie hacking and the business side of things
i also yap about whatever catches my attention so don't expect only tech tweets lol
quoted my projects below, check the replies if you're curious about what i've built
anyways, glad you're here ♡
Opus 5.5 vs Gemini 4 Hydrogen just exposed a pretty wild pricing gap
Same SVG challenge. Two cyclists. One clear winner on fine details
Opus 5.5: 60 minutes, $18
Gemini 4 Hydrogen: 15 minutes, $0.90
That's 20x the cost for Opus 😭
And here's the interesting part...
Hydrogen's result is still genuinely good. Opus looks more polished, but is that extra detail worth paying 20x more?
For production work, maybe. For rapid prototyping? That's a much harder sell
One test doesn't settle the model debate, but this is the kind of comparison developers actually need.
3
Gemini 4 Hydrogen reportedly beats Sonnet 5.5 on benchmarks
Then someone gave both models the SAME 3D koi pond challenge
And the results are making that benchmark win look a little awkward 😭
Hydrogen built a working jelly pond, but look at Sonnet's version...
The flowers, koi fish, lighting, composition, even the UI feels more polished
This is why I love real-world comparisons
A higher benchmark score doesn't automatically mean a better-looking result
Which one are you picking?
1
Google might have a serious weapon for rapid prototyping with Gemini 4 Hydrogen
Someone just generated this animated 3D Japanese teahouse in a SINGLE turn, reportedly in under a minute
No lengthy back-and-forth. Just one prompt at high reasoning effort
If this speed holds up across more complex projects, Claude and the other coding models have some competition coming
The real test is what happens when you ask it to modify the scene without breaking everything 👀
3
sou retweeted
Google's Gemini 4 lineup just produced a pretty weird result 😭
Someone tested Hydrogen, Carbon and Argon on the SAME stickman fighting game
And the pricing hierarchy didn't behave how you'd expect
Hydrogen: $2.46, 56 minutes
Argon: $4.65, 37 minutes
So Argon finished 19 minutes faster, while Hydrogen cost less but took significantly longer
And apparently Carbon was both faster AND cheaper than Hydrogen
Here's the interesting question...
If a cheaper model can finish your actual task faster, what exactly are you paying for with the higher tier?
Model selection is becoming its own engineering problem.
Gemini 4 Hydrogen reportedly beats Sonnet 5.5 on benchmarks
Then someone gave both models the SAME 3D koi pond challenge
And the results are making that benchmark win look a little awkward 😭
Hydrogen built a working jelly pond, but look at Sonnet's version...
The flowers, koi fish, lighting, composition, even the UI feels more polished
This is why I love real-world comparisons
A higher benchmark score doesn't automatically mean a better-looking result
Which one are you picking?
Google's Gemini 4 lineup just produced a pretty weird result 😭
Someone tested Hydrogen, Carbon and Argon on the SAME stickman fighting game
And the pricing hierarchy didn't behave how you'd expect
Hydrogen: $2.46, 56 minutes
Argon: $4.65, 37 minutes
So Argon finished 19 minutes faster, while Hydrogen cost less but took significantly longer
And apparently Carbon was both faster AND cheaper than Hydrogen
Here's the interesting question...
If a cheaper model can finish your actual task faster, what exactly are you paying for with the higher tier?
Model selection is becoming its own engineering problem.
Gemini 4 Hydrogen reportedly beats Sonnet 5.5 on benchmarks
Then someone gave both models the SAME 3D koi pond challenge
And the results are making that benchmark win look a little awkward 😭
Hydrogen built a working jelly pond, but look at Sonnet's version...
The flowers, koi fish, lighting, composition, even the UI feels more polished
This is why I love real-world comparisons
A higher benchmark score doesn't automatically mean a better-looking result
Which one are you picking?
Fable 5.2 just did something that would make artists VERY uncomfortable
Someone asked it to recreate a drawing, stroke by stroke
It drew the construction lines, eyes, facial details, everything
Then the creator asked if it had traced the original
Its reported response was the wild part:
"The original has uneven patches here. Mine doesn't."
Bro recreated the drawing and offered to FIX the reference 😭
3
People are calling Qwen 4 Max mid after testing it with ONE prompt 😭
Meanwhile, someone started with a traffic cone, a hula hoop and a few basic shapes
3 follow-ups later...
An entire floating London with Big Ben, the London Eye, skyscrapers, waterfalls and tiny floating islands
Look at the before and after. It's ridiculous
Makes you wonder how many "bad model" reviews are actually just unfinished conversations
1
Opus 5.5 vs Gemini 4 Hydrogen just exposed a pretty wild pricing gap
Same SVG challenge. Two cyclists. One clear winner on fine details
Opus 5.5: 60 minutes, $18
Gemini 4 Hydrogen: 15 minutes, $0.90
That's 20x the cost for Opus 😭
And here's the interesting part...
Hydrogen's result is still genuinely good. Opus looks more polished, but is that extra detail worth paying 20x more?
For production work, maybe. For rapid prototyping? That's a much harder sell
One test doesn't settle the model debate, but this is the kind of comparison developers actually need.
3
sou retweeted
Gemini 4 Hydrogen reportedly beats Sonnet 5.5 on benchmarks
Then someone gave both models the SAME 3D koi pond challenge
And the results are making that benchmark win look a little awkward 😭
Hydrogen built a working jelly pond, but look at Sonnet's version...
The flowers, koi fish, lighting, composition, even the UI feels more polished
This is why I love real-world comparisons
A higher benchmark score doesn't automatically mean a better-looking result
Which one are you picking?
Opus 5.5 vs Gemini 4 Hydrogen just exposed a pretty wild pricing gap
Same SVG challenge. Two cyclists. One clear winner on fine details
Opus 5.5: 60 minutes, $18
Gemini 4 Hydrogen: 15 minutes, $0.90
That's 20x the cost for Opus 😭
And here's the interesting part...
Hydrogen's result is still genuinely good. Opus looks more polished, but is that extra detail worth paying 20x more?
For production work, maybe. For rapid prototyping? That's a much harder sell
One test doesn't settle the model debate, but this is the kind of comparison developers actually need.
Gemini 4 Hydrogen reportedly beats Sonnet 5.5 on benchmarks
Then someone gave both models the SAME 3D koi pond challenge
And the results are making that benchmark win look a little awkward 😭
Hydrogen built a working jelly pond, but look at Sonnet's version...
The flowers, koi fish, lighting, composition, even the UI feels more polished
This is why I love real-world comparisons
A higher benchmark score doesn't automatically mean a better-looking result
Which one are you picking?
Opus 5.5 vs Gemini 4 Hydrogen just exposed a pretty wild pricing gap
Same SVG challenge. Two cyclists. One clear winner on fine details
Opus 5.5: 60 minutes, $18
Gemini 4 Hydrogen: 15 minutes, $0.90
That's 20x the cost for Opus 😭
And here's the interesting part...
Hydrogen's result is still genuinely good. Opus looks more polished, but is that extra detail worth paying 20x more?
For production work, maybe. For rapid prototyping? That's a much harder sell
One test doesn't settle the model debate, but this is the kind of comparison developers actually need.
Fable 5.2 just did something that would make artists VERY uncomfortable
Someone asked it to recreate a drawing, stroke by stroke
It drew the construction lines, eyes, facial details, everything
Then the creator asked if it had traced the original
Its reported response was the wild part:
"The original has uneven patches here. Mine doesn't."
Bro recreated the drawing and offered to FIX the reference 😭
5
Gemini 4 Hydrogen is making the whole "AI can't build real 3D worlds" argument look pretty shaky
Someone gave it a futuristic city to build in Three.js
ONE follow-up later, there's a flyable cyberpunk city with procedural skyscrapers, rain and flying traffic
And here's what gets me...
We're watching people go from a prompt to explorable worlds with barely any iteration
Imagine what happens when these models can keep building on the same world for days
Game development is about to get very interesting 👀
6
People are calling Qwen 4 Max mid after testing it with ONE prompt 😭
Meanwhile, someone started with a traffic cone, a hula hoop and a few basic shapes
3 follow-ups later...
An entire floating London with Big Ben, the London Eye, skyscrapers, waterfalls and tiny floating islands
Look at the before and after. It's ridiculous
Makes you wonder how many "bad model" reviews are actually just unfinished conversations
8
sou retweeted
Opus 5.5 vs Gemini 4 Hydrogen just exposed a pretty wild pricing gap
Same SVG challenge. Two cyclists. One clear winner on fine details
Opus 5.5: 60 minutes, $18
Gemini 4 Hydrogen: 15 minutes, $0.90
That's 20x the cost for Opus 😭
And here's the interesting part...
Hydrogen's result is still genuinely good. Opus looks more polished, but is that extra detail worth paying 20x more?
For production work, maybe. For rapid prototyping? That's a much harder sell
One test doesn't settle the model debate, but this is the kind of comparison developers actually need.
People are calling Qwen 4 Max mid after testing it with ONE prompt 😭
Meanwhile, someone started with a traffic cone, a hula hoop and a few basic shapes
3 follow-ups later...
An entire floating London with Big Ben, the London Eye, skyscrapers, waterfalls and tiny floating islands
Look at the before and after. It's ridiculous
Makes you wonder how many "bad model" reviews are actually just unfinished conversations
Opus 5.5 vs Gemini 4 Hydrogen just exposed a pretty wild pricing gap
Same SVG challenge. Two cyclists. One clear winner on fine details
Opus 5.5: 60 minutes, $18
Gemini 4 Hydrogen: 15 minutes, $0.90
That's 20x the cost for Opus 😭
And here's the interesting part...
Hydrogen's result is still genuinely good. Opus looks more polished, but is that extra detail worth paying 20x more?
For production work, maybe. For rapid prototyping? That's a much harder sell
One test doesn't settle the model debate, but this is the kind of comparison developers actually need.
People are calling Qwen 4 Max mid after testing it with ONE prompt 😭
Meanwhile, someone started with a traffic cone, a hula hoop and a few basic shapes
3 follow-ups later...
An entire floating London with Big Ben, the London Eye, skyscrapers, waterfalls and tiny floating islands
Look at the before and after. It's ridiculous
Makes you wonder how many "bad model" reviews are actually just unfinished conversations
Fable 5.2 just did something that would make artists VERY uncomfortable
Someone asked it to recreate a drawing, stroke by stroke
It drew the construction lines, eyes, facial details, everything
Then the creator asked if it had traced the original
Its reported response was the wild part:
"The original has uneven patches here. Mine doesn't."
Bro recreated the drawing and offered to FIX the reference 😭
7
Gemini 4 Hydrogen is making the whole "AI can't build real 3D worlds" argument look pretty shaky
Someone gave it a futuristic city to build in Three.js
ONE follow-up later, there's a flyable cyberpunk city with procedural skyscrapers, rain and flying traffic
And here's what gets me...
We're watching people go from a prompt to explorable worlds with barely any iteration
Imagine what happens when these models can keep building on the same world for days
Game development is about to get very interesting 👀
4