Pixel32Bench compares language models via 32x32 pixel generation

TheMoonMidas · x · 2026-08-23

A developer launched Pixel32Bench, a project to compare language models' ability to generate images pixel-by-pixel on a 32x32 canvas using a single prompt. The zero-shot benchmark uses a local Node.js harness to track time and API costs, enforcing a strict JSON schema output. It visualizes differences in model thinking, cost, and performance at a very low resolution.

Related event: Pixel32Bench Tests LLMs Drawing Mario on a 32x32 Grid(2 posts)→

Original post →

More from Models

Models channel →