AI/ML●●●Banger
A model-routing benchmark – the routers optimize the wrong axis
Proves cheap local models with RAG beat cold flagship models on citation accuracy.
Big BrainDark Horse
yadellopez
102d ago
Back to browse
SDF Pelicans on Bicycle
by AzatJ·Jul 18, 2026·2 points·0 comments
AI Analysis
●●●BangerBig BrainWizardry
Forces models to write ray marchers under 32KB when most demos are megabytes.
Strengths
- •Byte-exact prompt ensures reproducible results across different model runs.
- •GPU timer queries provide objective performance metrics beyond visual flair.
- •Strict constraints separate code golf skill from generic scaffolding reliance.
Weaknesses
- •Niche appeal limited to those caring about SDFs and model code generation.
- •Evaluation rubric relies on manual inspection for visual correctness.
Category
Target Audience
AI researchers and graphics programmers
Similar To
Simon Willison's Pelican challenge · Code Golf Stack Exchange
Similar Projects
AI/ML●●Solid
FretBench – I tested 14 LLMs on reading guitar tabs. Most failed
Clever benchmark exposing LLM tokenization weakness on ASCII art, but narrow domain.
Big BrainNiche Gem
jmcapra
104mo ago
AI/ML●●Solid
AdvertBench, ranking the ability of LLMs to create image ads
Human-voted ad benchmark as proxy for LLM tool-use ability.
Big BrainRabbit Hole
joegibbs
3028d ago
AI/ML●●Solid
Evaluating Local LLMs as language translators for my app
Local 18 GB Gemma ties frontier cloud on Afrikaans translation.
Big BrainNiche Gem
3stacks
421mo ago
AI/ML●●Solid
Reproducible benchmark – OpenAI charges 1.5x-3.3x more for non-English
Exposes 230% Arabic token tax that nobody talks about in pricing.
Dark HorseBig Brain
vfalbor
103mo ago
Developer Tools●●Solid
A reproducible React data grid benchmark with raw browser samples
Raw browser samples and deterministic fixtures make this benchmark actually reproducible.
Big BrainNiche Gem
vitashev
1018d ago