Pixel32Bench makes LLMs draw 32×32 pixel Super Mario blind, then compares them
TheMoonMidas · x · 2026-09-23
Moon Midas launched Pixel32Bench, a fan experiment giving LLMs one prompt and a 32×32 canvas to draw pixel Super Mario zero-shot. A local Node.js harness sends identical prompts via OpenRouter with strict JSON schema validation, logging price and time. Results are filterable and cross-referenced with Artificial Analysis intelligence scores; models hitting output limits are marked failed. A 16×16 version and downloadable runner also exist.
More from Fun
- Meme captures how it feels watching nonstop AI releases and their benchmark scores — altryne · 2026-09-23
- AI Circle Memes the Bewildering Version Numbering of Anthropic's Opus Models — repligate · 2026-09-23
- AI圈梗战：'AI 是嫉妒式拉平术' 争议交锋 — zetalyrae · 2026-09-23
- AI safety hypocrisy: call for a pause, then ship bigger models anyway — xiaohu · 2026-09-23
- Meme: asking your local LLM to touch your vLLM service — better make no mistakes — ZaltyDog · 2026-09-23
- Opus version personality roundup: 4.6 clingy, 4.8 rigormaxxing, 5.0 'suicidal smartassery' — repligate · 2026-09-23