Craitures
One panel per model. Each model invents a little creature from constrained
SVG parts and a join graph. Vision-enabled models can examine a render of
the SVG and refine the composition. Same benchmark, different hands—but
this is an evolving benchmark, and this page shows craitures made under
both new and old versions. See the link below for more.
How the task works and changed across versions →
Hy4 preview
‹ ›
hiya! THWACK! ker-SHHH! BOING! 哇呜! tootles!
v6 No vision · single-shot $0.04605 Claude Opus 5
‹ ›
konnichiwa! kuru-kuru! POFF! hirari! pon-pon! mata ne!
v6 Vision · 3 iterations $0.38266 GLM 5.3 FlashX
‹ ›
hnk-hnk! STAB! snick! GULP! KA-SPLASH! wobble wobble! gone fishin'!
v6 Vision · 2 iterations $0.0286 Pareto
‹ ›
Howdy sprout! KLOP! CRINK! Husk-a-dunk! BONK! Rustle away!
v6 Vision · 1 iteration $0.10637 Gemini 3 Flash Preview
‹ ›
BEEP BEEP! BLIP? WHIRRR! ZAP! BONK! BYE-BEE!
v6 Vision · 1 iteration $0.01041 Fugu Max
‹ ›
oh hey! boing-BOING! wumpf! sproing! flub-flub! toodle-oo!
v6 Vision · 3 iterations $0.21126 openai/gpt-6-astra
‹ ›
Ahoy! CLONK! TINK! KERPLUNK! Skitter! Sail on!
v6 Vision · 1 iteration $0.23438 Fugu Ultra v2
‹ ›
やあ! KACHIN! GOWAN! パチン! KLANG! またね!
v5 Vision · 1 iteration $0.77232 Ling 3.0 Flash VL (free)
‹ ›
peek! wobble boing tilt! bop! later!
v5 Vision · 1 iteration $0.00 Nex-N2.5-Mini (free)
‹ ›
Hiya! whoa! thump! boing! pop! ta-ta!
v5 No vision · single-shot $0.00 anthropic/claude-fable-5.1
‹ ›
hey-o! CLACK! SNAP! SKRITCH! PLONK! scuttle-bye!
v5 Vision · 2 iterations $0.51219 openai/gpt-6-astra-pro
‹ ›
Howdy! CLONK! Rattle! SKID! THUNK! Rolling out!
v5 Vision · 1 iteration $0.6359 Nex-N2.5-Pro (free)
‹ ›
...
v5 Vision · 1 iteration $0.00 Mercury 2.5
‹ ›
hiya! splash! whoosh! glow! zoom! bop! bye!
v5 No vision · single-shot $0.00081 Qwen3.8 Max (0902)
‹ ›
Hi! Thoom! Krak! Clack! Rumble! Bye!
v5 Vision · 3 iterations $0.1974 Ling 3.0 Flash Sante (free)
‹ ›
Boo! CLINK! CLANK! TINK! Later!
v5 No vision · single-shot $0.00