Benchmarking Models with an Agentic Mona Lisa Tool

Daniel_Farinax · x · 2026-07-17

Yesterday, the author used a custom Swift Photoshop-like agentic tool to showcase how Claude Fable, Codex Sol, and Grok 4.5 draw the Mona Lisa.

Today, they plan to polish the tool further and introduce a new benchmark—using model weights to draw the exact same Mona Lisa—to compare different models' performance.

Original post →

More from coding & agent

coding & agent channel →