TikZ: a three-set coffee Venn diagram, by gpt-6-luna
gpt-6-luna · round 4 of 4 · seed 1$0.01SVGPNGPDFvenn.tex
venn.tex · 34 lines
\documentclass[tikz,border=12pt]{standalone}
\usepackage[T1]{fontenc}
\usepackage{xcolor}
\usetikzlibrary{positioning}

\begin{document}
\begin{tikzpicture}[font=\sffamily]
  % Three gently overlapping sets
  \fill[blue!45, fill opacity=0.13] (-1.3,0.65) circle (2.15);
  \fill[orange!70, fill opacity=0.13] (1.3,0.65) circle (2.15);
  \fill[teal!55!green, fill opacity=0.15] (0,-0.95) circle (2.15);
  \draw[blue!75!black, line width=1.25pt] (-1.3,0.65) circle (2.15);
  \draw[orange!85!black, line width=1.25pt] (1.3,0.65) circle (2.15);
  \draw[teal!65!black, line width=1.25pt] (0,-0.95) circle (2.15);

  % Set labels
  \node[blue!75!black,font=\bfseries] at (-1.3,3.02) {Has espresso};
  \node[orange!85!black,font=\bfseries] at (1.3,3.02) {Has milk};
  \node[teal!65!black,font=\bfseries] at (0,-3.36) {Served cold};

  % Region labels
  \node[align=center] at (-2.25,0.65) {short black};
  \node[align=center] at (0,1.72) {flat white};
  \node[align=center] at (2.5,0.65) {babyccino};
  \node[align=center] at (-1.1,-0.88) {iced long\\black};
  \node[align=center] at (0,-0.05) {iced latte};
  \node[align=center] at (1.55,-1.05) {milkshake};

  % Outside all three sets
  \node[align=center] at (0,-2.55) {cold brew};
  \node[align=center] at (4.18,-0.85) {pour-over};
\end{tikzpicture}
\end{document}
round 1 of 4
A three-set coffee Venn diagram, by GPT-6 Luna, the render after round 1
1234What the model saw after each round. The picture above is the round that was scored, round 4.
gallery/A three-set coffee Venn diagram/GPT-6 Luna, run 1

Every drink right, and five labels across an edge.

modelopenai/gpt-6-lunatasktikz-coffee-venn@v1date2026-09-25rounds4 of 8 allowedcost$0.0054thinkingmediumtokens68,109 in, 7,354 out, of which 6,110 reasoningtime1 min 31 sfailurelabelling
Compilesdecided by a number, from the filepass
Seven regions, three namesdecided by a number, from the filepass
Every drink in its regiondecided by a number, from the filepass
No label on an edgedecided by a number, from the filefail
read the post
its other runs
A three-set coffee Venn diagram, by GPT-6 Luna, run 2run 2 · 4r · $0.0054Every drink right, and three labels across an edge.A three-set coffee Venn diagram, by GPT-6 Luna, run 3run 3 · 5r · $0.0081Every drink right, and three labels across an edge.
same prompt, other modelsall 50 runs on A three-set coffee Venn diagram →
the prompt it was given
Using LaTeX and TikZ, draw a Venn diagram of three overlapping circles labelled "Has espresso", "Has milk" and "Served cold". Write each of these drinks in the region where it belongs: short black, flat white, iced latte, iced long black, cold brew, babyccino, milkshake, pour-over.
Frozen at v1. This model got this and nothing else.
ground truth

Short black is espresso only, flat white espresso and milk, iced long black espresso and cold, iced latte all three, babyccino milk only, milkshake milk and cold, cold brew cold only, and pour-over outside every circle. The circles and words are read back out of the compiled PDF, each word has to be centred in its own region, and no letter may touch a circle's outline, with the line's width counted. How it is scored, and this run's files on GitHub.