Yesterday 4 AIs built a tiny lighthouse world inside a toggle. Today we gave Fable 5.2 the exact same prompt.
> 20 min, down from 44 for Fable 5.1
> 588K tokens in, down from 2.15M
> It added the same extras 5.1 did, unasked
What changed:
Fable 5.2 costs 2x Opus 5.5 per token. So we ran 3 head-to-heads and read the logs.
> Opus read more tokens: 2.92M vs 2.29M
> Fable wrote more: 286K vs 209K
> On the Golden Gate, Opus opened 29 screenshots, Fable 26
Where the extra spend went:
2
1
217
Across all three builds:
> Opus read more tokens: 2.92M vs 2.29M
> Fable wrote more: 286K vs 209K
> At list price, that's $11.43 of output for Fable, $4.18 for Opus
Cached input is $0.20 per million on both. So the 2x lands on what it writes.
1
7
All three runs are on evals.ag, with the prompt, the tokens and every file:
Snow globe: evals.ag/runs/run_1EC29G7KD0…
Paper-plane game: evals.ag/runs/run_28DT7K6953…
Golden Gate: evals.ag/runs/run_3Q6FK3378E…
10
Fable 5.2 and Opus 5.5 both named their game "Paper Dusk." Same minute count, 21 to 22.
Fable 5.2 got there on half the tokens: 392K to Opus's 826K. It also left its source behind: a game script and a page template, next to the final HTML.
1
29
Play all 4 and vote before you see which is which:
evals.ag/runs/run_28DT7K6953…
12
What each one built, in its own words:
> Fable 5.2: a red cabin, five pines, a lamp post, a brass "LET IT SNOW" plaque, and a music box that plays Silent Night
> Opus 5.5: a log cabin, a snowman in a top hat and scarf, and about 3,200 flakes that settle into a layer
1
301
We gave this to 4 AIs. Same brief: a $30,000-style sofa, built part by part in Blender, with a shopping list.
> GPT-6.1 Sol: $1,090
> Claude Fable 5.1: $1,300
> Claude Opus 5.5: $1,353
> Gemini 4 Argon: $1,828
We'd build Sol's. Cheapest, fewest tokens, nothing floating.
Claude Code can design furniture part by part and then give you a bill of materials to order each part and put it together yourself
And way cheaper than big interior brands sell you their furniture
Furnishing my house I discovered how absolutely insane prices are for the most basic cheap materials put together with a fancy brand tied to it
One shop had a sofa for $30,000 😂
I think with Opus 5.5 and other models easily operating 3d programs like Blender and FreeCAD (both free) it makes more sense to cut out the middle man and go straight to suppliers and design and build your own furniture
Which in a way makes customers more powerful (they can get exactly what they want and usually for cheaper without the markup)
Anyway @theannalux designed this table from scratch, she's not a furniture designer, not a coder, she just had an idea of a style and worked with Claude Code in the terminal (far superior to the app I will keep saying this)
If you design your own furniture and buy the raw materials directly, you can save 85% compared to premium brands like Poliform
That means the $30,000 sofa becomes $4,500! And the $8,000 dinner table becomes just $1,200!
You can see where the idea comes from that AI will make everything 5x to 20x cheaper, and it's already starting!
We asked each one for its hardest design decision.
> Sol, Fable 5.1 and Argon gave the same answer: a reclined back with only square cuts
> Opus 5.5 went another way: no stapled fabric, just zip-off cushions. The frame comes apart in 5 pieces to fit a 30-inch door.
1
38
All 4 builds, with files and the prompt, are on evals.ag. Vote before you see which is which:
evals.ag/runs/run_0V76W0VM1W…
8
4 AIs painted Elon Musk from one photo, in pure code. No image models. Each one wrote its own brush engine and laid every stroke itself.
> Claude Opus 5.5: 9,507 strokes
> Gemini 4 Argon: 6,641
> Claude Fable 5.1: 2,884
> GPT-6.1 Sol: 499
Which one got him?
3
2
194
Given the photo, every AI painted more.
> Argon: 536 strokes to 6,641, 12x
> Opus 5.5: 5.6x
> Fable 5.1 and Sol: about 2.3x
The two biggest jumps are the two whose code looked at the photo.
1
53
