Pelican SVG Benchmark

A complex generative vector art benchmark testing LLM reasoning and spatial awareness.

PlayableOpen source

Description

A vector art benchmark used to test the reasoning capabilities of local LLMs. This iteration, produced by Qwen 3.8 27B, successfully rendered a detailed pelican riding a bicycle with correct frame geometry, legs on both sides, and a distinct pouch, requiring 22,276 reasoning tokens to complete at 'xhigh' reasoning effort.

Descriptions, tags, and model credits may be AI-generated or inferred from public sources and can be incomplete or wrong. Learn more.

Details

Why this is interesting: Demonstrates a local 27B model's ability to handle complex spatial reasoning and vector art generation through a…
Attributed to: @simonw
Built with: Qwen3.8-27B Grok 4.5 Seedream 4.5 GPT-5
Links: simonwillison.net
View on YouTube
View GitHub repo
Tags: #svg #benchmark #vector-art #reasoning

Discovered: YouTube
First posted: YouTube · Aug 17, 2026
Listed: Added 55d ago
Build: Strong
Models: Strong

Prompt

draw an svg of a circle

Source